Maria SorokinapersName, Peter MerseburgerpersName, Kohulan RajanpersName, Mehmet Aziz YirikpersName, Christoph SteinbeckpersName (2021date). “COCONUT online: Collection of Open Natural Products database.” Journal of Cheminformaticstitle·j.
Object of study — the aggregation, curation, annotation, storage, and online retrieval of open natural-product structures and provenance
Research question — whether distributed open natural-product collections can be unified into a general, freely accessible database that supports chemical and programmatic search
Contribution — the COCONUTterm aggregated datasetterm, its chemical and provenance curation pipeline, and a searchable, downloadable, containerized web database with a REST APIterm
Cross-disciplinary analysis
content-tags & facets:
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | fragmented natural-product knowledge, heterogeneous molecular representations, incomplete provenance metadata, and chemical retrieval across a large unified collection — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | |
| claim-type | existence — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | |
| method — evidence | — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | |
| object of study — evidence | — “COCONUT is assembled from a large number of chemical data sources (Table 1 ), from which NPs have been thoroughly extracted, curated, processed and annotated. The resulting NPs collection is presented within the full-fledged chemical database developed specially for this purpose (Fig. 1 ). Data curation and processing steps, together with the construction of the web interface and the description of available features are described below.” | |
| research question — evidence | — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | |
| contribution — evidence | — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | |
| mechanism | source aggregation Extracts, curates, processes, and annotates natural products from many chemical sources. — “COCONUT is assembled from a large number of chemical data sources (Table 1 ), from which NPs have been thoroughly extracted, curated, processed and annotated. The resulting NPs collection is presented within the full-fledged chemical database developed specially for this purpose (Fig. 1 ). Data curation and processing steps, together with the construction of the web interface and the description of available features are described below.” | mechanisms[0] |
| mechanism | structural quality control Checks size, connectivity, pseudo-atoms, hydrogens, bonds, and valence before registration. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | mechanisms[1] |
| mechanism | ChEMBL standardization Normalizes tautomer and ionization states using an established chemical-curation pipeline. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | mechanisms[2] |
| mechanism | InChIKey unification Collapses records across sources without relying on inconsistently supplied stereochemistry. — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | mechanisms[3] |
| mechanism | automatic metadata curation Retrieves names, synonyms, cross-references, literature identifiers, and taxonomy. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | mechanisms[4] |
| mechanism | ClassyFire Assigns hierarchical chemical classes that enable class-based grouping. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | mechanisms[5] |
| mechanism | five-star annotation score Summarizes the presence of verified names, taxonomy, literature, and trusted provenance. — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | mechanisms[6] |
| mechanism | name resolution Combines original names, InChI searches in major databases, and computed IUPAC names. — “NPs common names in COCONUT have been retrieved, when available, from their databases of origin. The remaining NPs have been searched by InChI in major chemical databases (PubChem, ChEMBL and ChEBI) and common names and synonyms were retrieved when occurences the compound were found. Additionally to this, IUPAC names were computed for all COCONUT NPs using ChemAxon’s MolCovert [ 20 ], to add nomenclature homogeneity to the dataset. Furthemore, the IUPAC names are used as the main NP name when no official chemical name has been found nor in the original sources, nor by searching big chemical databases. Therefore, all NPs in COCONUT have at least one molecular name.” | mechanisms[7] |
| mechanism | Sugar Removal Utility Detects sugar moieties and provides a deglycosylated representation. — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | mechanisms[8] |
| mechanism | SuperNatural II provenance screen Retains source-only molecules only when at least one natural-product plausibility test passes. — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | mechanisms[9] |
| mechanism | MongoDB document storage Stores complex records and supplies indexing, text search, and aggregation functions. — “All COCONUT data is stored with MongoDB, a cross-platform document-oriented NoSQL database program. The smallest unit in MongoDB is a document, composed of key and value pairs that are similar to JSON objects. Documents of the same nature are organized in collections, which are the equivalent of the SQL-based databases tables. MongoDB is particularly adapted to big and complex data, supports multiple indexing, including text indexing allowing enhanced text search in text-indexed fields and contains a wide range of in-build search and analysis functions.” | mechanisms[10] |
| mechanism | regular-expression routing Identifies a simple-search input type before querying the corresponding database field. — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | mechanisms[11] |
| mechanism | fingerprint prefilter Uses all query bits to select candidate structures before exact pattern matching. — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | mechanisms[12] |
| mechanism | CDK pattern matching Confirms substructure candidates with Ullmann, Vento-Foggia, or depth-first matching. — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | mechanisms[13] |
| mechanism | inverted fingerprint index Maps PubChem fingerprint bits to COCONUT identifiers for LSH-based similarity retrieval. — “Similarity search with MongoDB was implemented following the excellent ChEBML blog post on LSH-based similarity search in MongoDB [ 32 ] and adapted to Java, Kotlin and Spring data. In this approach, the MongoDB aggregation framework is used to perform inverted indexing search against PubChem fingerprints stored in a separate table and referencing COCONUT identifiers that contain the molecular features encoded by each bit.” | mechanisms[14] |
| mechanism | REST API Exposes programmatic queries for integrating COCONUT into external workflows. — “A REST API has been developed for COCONUT online in order to permit programmatic querying and facilitate its integration in workflows. It relies on Kotlin API functionalities and it’s usage, together with some examples, is described in detail in the documentation section of the website ( https://coconut.naturalproducts.net/documentation ).” | mechanisms[15] |
| contrasts-with | specialized, commercial, unavailable, or sparsely annotated natural-product databases — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | claim_relations[0] |
| builds-on | 53 open data sources and manually collected literature sets — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | claim_relations[1] |
| builds-on | the ChEMBL chemical structure curation pipeline — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | claim_relations[2] |
| builds-on | LSH-based similarity search in MongoDB — “Similarity search with MongoDB was implemented following the excellent ChEBML blog post on LSH-based similarity search in MongoDB [ 32 ] and adapted to Java, Kotlin and Spring data. In this approach, the MongoDB aggregation framework is used to perform inverted indexing search against PubChem fingerprints stored in a separate table and referencing COCONUT identifiers that contain the molecular features encoded by each bit.” | claim_relations[3] |
| sense | natural product as a small molecule produced by a living organism — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | senses[0] |
| sense | COCONUT as both an aggregated dataset and an online access system — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | senses[1] |
| sense | generalist database as an integrator across thematic collections — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | senses[2] |
| sense | open access as no-login browsing plus downloadable data and unrestricted code — “The COCONUT database is free and open to all users and there is no login required to access it. Its web interface allows diverse simple searches (e.g. by molecule name, InChI, InChI key, SMILES, drawn structure, molecular formula), advanced search by molecular features, together with substructure and similarity searches. Users can also download the whole dataset or search results in different formats. The database can be queried programmatically via a REST API, which facilitates COCONUT integration in workflows. The web interface, the back-end and the database are deployed as Docker containers, making it easily portable for hosting other sets of NPs and to be deployed on local installations.” | senses[3] |
| sense | flat natural-product identity as InChIKey equality without stereochemistry — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | senses[4] |
| sense | stereochemistry as preserved source information rather than the unification key — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | senses[5] |
| sense | annotation quality as metadata completeness and trusted provenance rather than chemical validity alone — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | senses[6] |
| sense | computed IUPAC name as a fallback when no official or common name is found — “NPs common names in COCONUT have been retrieved, when available, from their databases of origin. The remaining NPs have been searched by InChI in major chemical databases (PubChem, ChEMBL and ChEBI) and common names and synonyms were retrieved when occurences the compound were found. Additionally to this, IUPAC names were computed for all COCONUT NPs using ChemAxon’s MolCovert [ 20 ], to add nomenclature homogeneity to the dataset. Furthemore, the IUPAC names are used as the main NP name when no official chemical name has been found nor in the original sources, nor by searching big chemical databases. Therefore, all NPs in COCONUT have at least one molecular name.” | senses[7] |
| sense | NP-likeness as a plausibility feature rather than proof of natural provenance — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | senses[8] |
| sense | marine as an environmental grouping rather than a monoclade taxon — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | senses[9] |
| sense | geographic provenance as occurrence or collection location rather than necessarily evolutionary origin — “The geographic location of the collection or the natural presence of the NP-producing organism is a piece of information that is even more difficult to obtain. Nowadays, a range of organisms, and in particular plants, can be found in different parts of the planet due to globalisation and their success in human consumption (e.g. garlic, tomatoes, curcuma or ginger). It is, therefore, difficult, if not impossible, to determine their original provenance. Also, the geographical information is often omitted in literature and most NPs databases. When available, the geographical provenance is stored in the MongoDB dump of COCONUT, but not displayed on the website.” | senses[10] |
| sense | simple search as input-type dispatch across heterogeneous identifier fields — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | senses[11] |
| sense | substructure match as fingerprint screening followed by exact graph confirmation — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | senses[12] |
| sense | REST API as workflow integration rather than only interactive web access — “A REST API has been developed for COCONUT online in order to permit programmatic querying and facilitate its integration in workflows. It relies on Kotlin API functionalities and it’s usage, together with some examples, is described in detail in the documentation section of the website ( https://coconut.naturalproducts.net/documentation ).” | senses[13] |
| control | structure-size, connectivity, atom, hydrogen, bond, and valence checks — held — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” — rules out invalid or unsupported external structures entering registration unchecked — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | controls[0] |
| control | stereochemistry-free identity plus preservation of original stereochemical structures — held — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” — rules out inconsistent source stereochemistry preventing flat-record unification while discarding all source detail — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | controls[1] |
| control | additional filtering only for SuperNatural II-exclusive records — held — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” — rules out questionable source-specific natural-product assignments contaminating the integrated collection — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | controls[2] |
| control | fingerprint all-bits-set prefilter followed by exact CDK graph matching — held — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” — rules out treating fingerprint containment alone as a confirmed substructure — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | controls[3] |
| note | specialized, commercial, unavailable, or sparsely annotated natural-product databases — The contribution is motivated against the fragmented database landscape. — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | claim_relations[0] |
| weight | specialized, commercial, unavailable, or sparsely annotated natural-product databases — 0.95 — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | claim_relations[0] |
| note | 53 open data sources and manually collected literature sets — The new collection directly aggregates these sources. — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | claim_relations[1] |
| weight | 53 open data sources and manually collected literature sets — 0.95 — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | claim_relations[1] |
| note | the ChEMBL chemical structure curation pipeline — Tautomer and ionization-state standardization explicitly follows ChEMBL. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | claim_relations[2] |
| weight | the ChEMBL chemical structure curation pipeline — 0.95 — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | claim_relations[2] |
| note | LSH-based similarity search in MongoDB — The implementation adapts the cited MongoDB similarity-search design. — “Similarity search with MongoDB was implemented following the excellent ChEBML blog post on LSH-based similarity search in MongoDB [ 32 ] and adapted to Java, Kotlin and Spring data. In this approach, the MongoDB aggregation framework is used to perform inverted indexing search against PubChem fingerprints stored in a separate table and referencing COCONUT identifiers that contain the molecular features encoded by each bit.” | claim_relations[3] |
| weight | LSH-based similarity search in MongoDB — 0.95 — “Similarity search with MongoDB was implemented following the excellent ChEBML blog post on LSH-based similarity search in MongoDB [ 32 ] and adapted to Java, Kotlin and Spring data. In this approach, the MongoDB aggregation framework is used to perform inverted indexing search against PubChem fingerprints stored in a separate table and referencing COCONUT identifiers that contain the molecular features encoded by each bit.” | claim_relations[3] |
| note | MongoDB —is-a→ document-oriented NoSQL database program — Direct categorical definition. — “All COCONUT data is stored with MongoDB, a cross-platform document-oriented NoSQL database program. The smallest unit in MongoDB is a document, composed of key and value pairs that are similar to JSON objects. Documents of the same nature are organized in collections, which are the equivalent of the SQL-based databases tables. MongoDB is particularly adapted to big and complex data, supports multiple indexing, including text indexing allowing enhanced text search in text-indexed fields and contains a wide range of in-build search and analysis functions.” | concept_edges[0] |
| relation | MongoDB —is-a→ document-oriented NoSQL database program — MongoDB is described as a document-oriented NoSQL database program. — “All COCONUT data is stored with MongoDB, a cross-platform document-oriented NoSQL database program. The smallest unit in MongoDB is a document, composed of key and value pairs that are similar to JSON objects. Documents of the same nature are organized in collections, which are the equivalent of the SQL-based databases tables. MongoDB is particularly adapted to big and complex data, supports multiple indexing, including text indexing allowing enhanced text search in text-indexed fields and contains a wide range of in-build search and analysis functions.” | concept_edges[0] |
| note | SourceNaturalProduct —part-of→ COCONUT — Explicit component membership. — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | concept_edges[1] |
| relation | SourceNaturalProduct —part-of→ COCONUT — SourceNaturalProduct is one of COCONUT's two major database collections. — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | concept_edges[1] |
| note | UniqueNaturalProduct —part-of→ COCONUT — Explicit component membership. — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | concept_edges[2] |
| relation | UniqueNaturalProduct —part-of→ COCONUT — UniqueNaturalProduct is one of COCONUT's two major database collections. — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | concept_edges[2] |
| note | NPclassifier —is-a→ deep neural network-based structural classification tool — Direct categorical definition. — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | concept_edges[3] |
| relation | NPclassifier —is-a→ deep neural network-based structural classification tool — NPclassifier is explicitly described as a deep neural network-based structural classification tool. — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | concept_edges[3] |
| note | deepSMILES —is-a→ adaptation of SMILES — Direct categorical definition. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concept_edges[4] |
| relation | deepSMILES —is-a→ adaptation of SMILES — DeepSMILES is explicitly described as an adaptation of SMILES. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concept_edges[4] |
| note | annotation level —is-a→ 5-star-based system — Direct categorical definition. — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | concept_edges[5] |
| relation | annotation level —is-a→ 5-star-based system — The annotation level is a 5-star-based system. — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | concept_edges[5] |
| outcome | control — held — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | controls[0] |
| outcome | control — held — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | controls[1] |
| outcome | control — held — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | controls[2] |
| outcome | control — held — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | controls[3] |
| relation | distributed open data sources → COCONUT — COCONUT aggregates natural-product records from many open chemical sources. — “COCONUT is assembled from a large number of chemical data sources (Table 1 ), from which NPs have been thoroughly extracted, curated, processed and annotated. The resulting NPs collection is presented within the full-fledged chemical database developed specially for this purpose (Fig. 1 ). Data curation and processing steps, together with the construction of the web interface and the description of available features are described below.” | unhoused_relations[0] |
| relation | structural quality control → registered molecules — Every collected molecule passes structural checks before registration. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | unhoused_relations[1] |
| relation | ChEMBL chemical structure curation pipeline → tautomers and ionisation states — COCONUT follows the ChEMBL pipeline to standardize tautomer and ionization states. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | unhoused_relations[2] |
| relation | InChI keys without stereochemistry → unified natural-product records — Records from different sources are unified under stereochemistry-free InChIKey identity. — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | unhoused_relations[3] |
| relation | stereochemistry-free unification → stereoisomer-specific biological activity — The necessary unification policy does not preserve stereoisomer distinctions in flat records. — “The authors are well aware that different stereoisomers of a compound can have very different biological activity. The procedure described above was a necessary step to create a unified resource out of distributed databases of varying quality. Further curation will gradually improve stereochemical assignments and linkage to original source articles.” | unhoused_relations[4] |
| relation | automatic metadata curation → names, cross-references, literature, and taxonomy — Automatic curation retrieves and corrects multiple metadata classes. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | unhoused_relations[5] |
| weight | automatic metadata curation → names, cross-references, literature, and taxonomy — 0.95 — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | unhoused_relations[5] |
| relation | trusted data provenance → annotation stars — Trusted source membership contributes to a higher annotation-quality score. — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | unhoused_relations[6] |
| weight | trusted data provenance → annotation stars — 0.95 — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | unhoused_relations[6] |
| relation | Sugar Removal Utility → deglycosylated structure representation — Sugar-moiety manipulation supplies a deglycosylated representation. — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | unhoused_relations[7] |
| relation | additional SuperNatural II quality control → 24880 source-only molecules — The additional screen removed molecules lacking any accepted natural-product signal. — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | unhoused_relations[8] |
| relation | missing source metadata → incomplete natural-product provenance — Omitted organism, geography, and literature links leave most records incompletely annotated. — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | unhoused_relations[9] |
| weight | missing source metadata → incomplete natural-product provenance — 0.95 — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | unhoused_relations[9] |
| relation | COCONUT dataset scale → manual curation — The amount of data makes exhaustive manual curation infeasible. — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | unhoused_relations[10] |
| weight | COCONUT dataset scale → manual curation — 0.95 — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | unhoused_relations[10] |
| relation | symbionts and microbiomes → organism attribution — Compounds made by symbionts can be mistakenly assigned to the sampled host organism. — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | unhoused_relations[11] |
| weight | symbionts and microbiomes → organism attribution — 0.95 — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | unhoused_relations[11] |
| relation | globalisation and omitted location data → original geographic provenance — Organism movement and missing literature metadata make original provenance difficult to determine. — “The geographic location of the collection or the natural presence of the NP-producing organism is a piece of information that is even more difficult to obtain. Nowadays, a range of organisms, and in particular plants, can be found in different parts of the planet due to globalisation and their success in human consumption (e.g. garlic, tomatoes, curcuma or ginger). It is, therefore, difficult, if not impossible, to determine their original provenance. Also, the geographical information is often omitted in literature and most NPs databases. When available, the geographical provenance is stored in the MongoDB dump of COCONUT, but not displayed on the website.” | unhoused_relations[12] |
| weight | globalisation and omitted location data → original geographic provenance — 0.95 — “The geographic location of the collection or the natural presence of the NP-producing organism is a piece of information that is even more difficult to obtain. Nowadays, a range of organisms, and in particular plants, can be found in different parts of the planet due to globalisation and their success in human consumption (e.g. garlic, tomatoes, curcuma or ginger). It is, therefore, difficult, if not impossible, to determine their original provenance. Also, the geographical information is often omitted in literature and most NPs databases. When available, the geographical provenance is stored in the MongoDB dump of COCONUT, but not displayed on the website.” | unhoused_relations[12] |
| relation | simple searches → COCONUT — COCONUT exposes simple searches over names and chemical identifiers. — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | unhoused_relations[13] |
| relation | advanced search → COCONUT — COCONUT exposes advanced search over molecular properties. — “The advanced search supports searching for NPs in COCONUT according to a range of parameters, such as molecular formula, molecular descriptor values, number of rings, type of sugar moieties present in them, etc.” | unhoused_relations[14] |
| relation | PubChem fingerprint prefilter → exact substructure candidates — The all-bits-set query selects candidates that are then confirmed by graph matching. — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | unhoused_relations[15] |
| relation | REST API → external workflows — Programmatic querying facilitates integration of COCONUT in workflows. — “A REST API has been developed for COCONUT online in order to permit programmatic querying and facilitate its integration in workflows. It relies on Kotlin API functionalities and it’s usage, together with some examples, is described in detail in the documentation section of the website ( https://coconut.naturalproducts.net/documentation ).” | unhoused_relations[16] |
| weight | REST API → external workflows — 0.95 — “A REST API has been developed for COCONUT online in order to permit programmatic querying and facilitate its integration in workflows. It relies on Kotlin API functionalities and it’s usage, together with some examples, is described in detail in the documentation section of the website ( https://coconut.naturalproducts.net/documentation ).” | unhoused_relations[16] |
| relation | Dockerisation → local and cloud deployment — Containerization permits rapid deployment on local servers and cloud infrastructure. — “COCONUT web interface, back-end and database are entirely Dockerised, allowing a quick and easy deployment on local servers and cloud. All the code, for both front-end and back-end, is available on GitHub ( https://github.com/mSorok/NaturalProductsOnline ).” | unhoused_relations[17] |
| weight | Dockerisation → local and cloud deployment — 0.95 — “COCONUT web interface, back-end and database are entirely Dockerised, allowing a quick and easy deployment on local servers and cloud. All the code, for both front-end and back-end, is available on GitHub ( https://github.com/mSorok/NaturalProductsOnline ).” | unhoused_relations[17] |
| no type for this | distributed open data sources → COCONUT — COCONUT aggregates natural-product records from many open chemical sources. — “COCONUT is assembled from a large number of chemical data sources (Table 1 ), from which NPs have been thoroughly extracted, curated, processed and annotated. The resulting NPs collection is presented within the full-fledged chemical database developed specially for this purpose (Fig. 1 ). Data curation and processing steps, together with the construction of the web interface and the description of available features are described below.” | unhoused_relations[0] |
| no type for this | structural quality control → registered molecules — Every collected molecule passes structural checks before registration. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | unhoused_relations[1] |
| no type for this | ChEMBL chemical structure curation pipeline → tautomers and ionisation states — COCONUT follows the ChEMBL pipeline to standardize tautomer and ionization states. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | unhoused_relations[2] |
| no type for this | InChI keys without stereochemistry → unified natural-product records — Records from different sources are unified under stereochemistry-free InChIKey identity. — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | unhoused_relations[3] |
| no type for this | stereochemistry-free unification → stereoisomer-specific biological activity — The necessary unification policy does not preserve stereoisomer distinctions in flat records. — “The authors are well aware that different stereoisomers of a compound can have very different biological activity. The procedure described above was a necessary step to create a unified resource out of distributed databases of varying quality. Further curation will gradually improve stereochemical assignments and linkage to original source articles.” | unhoused_relations[4] |
| no type for this | automatic metadata curation → names, cross-references, literature, and taxonomy — Automatic curation retrieves and corrects multiple metadata classes. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | unhoused_relations[5] |
| no type for this | trusted data provenance → annotation stars — Trusted source membership contributes to a higher annotation-quality score. — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | unhoused_relations[6] |
| no type for this | Sugar Removal Utility → deglycosylated structure representation — Sugar-moiety manipulation supplies a deglycosylated representation. — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | unhoused_relations[7] |
| no type for this | additional SuperNatural II quality control → 24880 source-only molecules — The additional screen removed molecules lacking any accepted natural-product signal. — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | unhoused_relations[8] |
| no type for this | missing source metadata → incomplete natural-product provenance — Omitted organism, geography, and literature links leave most records incompletely annotated. — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | unhoused_relations[9] |
| no type for this | COCONUT dataset scale → manual curation — The amount of data makes exhaustive manual curation infeasible. — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | unhoused_relations[10] |
| no type for this | symbionts and microbiomes → organism attribution — Compounds made by symbionts can be mistakenly assigned to the sampled host organism. — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | unhoused_relations[11] |
| no type for this | globalisation and omitted location data → original geographic provenance — Organism movement and missing literature metadata make original provenance difficult to determine. — “The geographic location of the collection or the natural presence of the NP-producing organism is a piece of information that is even more difficult to obtain. Nowadays, a range of organisms, and in particular plants, can be found in different parts of the planet due to globalisation and their success in human consumption (e.g. garlic, tomatoes, curcuma or ginger). It is, therefore, difficult, if not impossible, to determine their original provenance. Also, the geographical information is often omitted in literature and most NPs databases. When available, the geographical provenance is stored in the MongoDB dump of COCONUT, but not displayed on the website.” | unhoused_relations[12] |
| no type for this | simple searches → COCONUT — COCONUT exposes simple searches over names and chemical identifiers. — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | unhoused_relations[13] |
| no type for this | advanced search → COCONUT — COCONUT exposes advanced search over molecular properties. — “The advanced search supports searching for NPs in COCONUT according to a range of parameters, such as molecular formula, molecular descriptor values, number of rings, type of sugar moieties present in them, etc.” | unhoused_relations[14] |
| no type for this | PubChem fingerprint prefilter → exact substructure candidates — The all-bits-set query selects candidates that are then confirmed by graph matching. — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | unhoused_relations[15] |
| no type for this | REST API → external workflows — Programmatic querying facilitates integration of COCONUT in workflows. — “A REST API has been developed for COCONUT online in order to permit programmatic querying and facilitate its integration in workflows. It relies on Kotlin API functionalities and it’s usage, together with some examples, is described in detail in the documentation section of the website ( https://coconut.naturalproducts.net/documentation ).” | unhoused_relations[16] |
| no type for this | Dockerisation → local and cloud deployment — Containerization permits rapid deployment on local servers and cloud infrastructure. — “COCONUT web interface, back-end and database are entirely Dockerised, allowing a quick and easy deployment on local servers and cloud. All the code, for both front-end and back-end, is available on GitHub ( https://github.com/mSorok/NaturalProductsOnline ).” | unhoused_relations[17] |
| limitation | Many existing natural-product databases are offline, commercial, narrowly specialized, or poorly annotated. — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | limitations[0] |
| limitation | Source databases vary in whether and how stereochemistry is represented. — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | limitations[1] |
| limitation | Unifying without stereochemistry can merge stereoisomers with very different biological activities. — “The authors are well aware that different stereoisomers of a compound can have very different biological activity. The procedure described above was a necessary step to create a unified resource out of distributed databases of varying quality. Further curation will gradually improve stereochemical assignments and linkage to original source articles.” | limitations[2] |
| limitation | The source-integration procedure reflects distributed databases of varying quality. — “The authors are well aware that different stereoisomers of a compound can have very different biological activity. The procedure described above was a necessary step to create a unified resource out of distributed databases of varying quality. Further curation will gradually improve stereochemical assignments and linkage to original source articles.” | limitations[3] |
| limitation | Only a selected subset of the 73 computed fields is displayed in the web interface. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | limitations[4] |
| limitation | Common names, taxonomic provenance, literature references, and trusted sources are absent from low-quality records. — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | limitations[5] |
| limitation | One contributing source required additional screening because its records had uncertain natural-product status. — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | limitations[6] |
| limitation | Ring counts can be ambiguous for fused and condensed ring systems. — “Counting rings in a molecule can become a complex task, as the outer perimeter of two fused rings can be counted as one big ring. With more condensed rings, the number of fused ring perimeters (aka as the set of all rings) can grow steeply. In Fig. 2 , only the minimal ring count (the minimal cycle base) is represented.” | limitations[7] |
| limitation | Most records still lack links among original publications, structures, organisms, and geography. — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | limitations[8] |
| limitation | The scale of COCONUT makes comprehensive manual curation infeasible. — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | limitations[9] |
| limitation | Organism provenance may be misassigned when symbionts or microbiomes produce an isolated molecule. — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | limitations[10] |
| limitation | Global movement of organisms and omission in literature make original geographic provenance difficult or impossible to infer. — “The geographic location of the collection or the natural presence of the NP-producing organism is a piece of information that is even more difficult to obtain. Nowadays, a range of organisms, and in particular plants, can be found in different parts of the planet due to globalisation and their success in human consumption (e.g. garlic, tomatoes, curcuma or ginger). It is, therefore, difficult, if not impossible, to determine their original provenance. Also, the geographical information is often omitted in literature and most NPs databases. When available, the geographical provenance is stored in the MongoDB dump of COCONUT, but not displayed on the website.” | limitations[11] |
| limitation | Geographic coverage is biased by uneven research attention and dataset sizes. — “For NPs where geographical information is available, it appears that most of them are produced by organisms that have been isolated in Asia (Fig. 4 ). This bias is introduced by the intensive study by scientists of the traditional Chinese and Indian medicines and by the big efforts in isolation and elucidation of NPs from medicinal plants. NPs from the African continent are also well represented in COCONUT (Fig. 4 ), mainly due to the scientific interest in African traditional medicines and African biodiversity. There is, for now, no data from the biodiversity of the Australian continent, and only very little data for NPs isolated from endemic European organisms. NPs from the Americas are mainly extracted and solved while Brazilian and Mexican biodiversity exploration. Only a few NPs are present in more than one continent, mainly in Asia and Africa, and the overlap values are biased by the very different NPs set sizes between the different continents.” | limitations[12] |
| limitation | Further annotation and functionality development remain necessary, especially for organisms, geography, and literature links. — “The final aim of COCONUT is to provide to the scientific community NPs structures and their provenance, i.e.. organisms that synthesize them and geographic location of the latter. However, a lot of data curation, in particular using new generation deep learning-based methods of extracting information from publications and books, together with website functionalities developments are still need to be done for COCONUT, but the database as it is now is already an important tool to facilitate NPs and medicinal chemistry research.” | limitations[13] |
Content tags
content-observations:
finding · quantity · variable · definition · key-term — grounded in the full text| finding | COCONUT aggregates elucidated and predicted natural products from open sources and provides a web interface for browsing, searching, and downloading them. — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | content_tags.findings[0] |
| finding | The database addresses a landscape of more than 120 published natural-product databases, including unavailable, commercial, specialized, and sparsely annotated resources. — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | content_tags.findings[1] |
| finding | COCONUT requires no login and supports simple, advanced, substructure, similarity, and programmatic searches. — “The COCONUT database is free and open to all users and there is no login required to access it. Its web interface allows diverse simple searches (e.g. by molecule name, InChI, InChI key, SMILES, drawn structure, molecular formula), advanced search by molecular features, together with substructure and similarity searches. Users can also download the whole dataset or search results in different formats. The database can be queried programmatically via a REST API, which facilitates COCONUT integration in workflows. The web interface, the back-end and the database are deployed as Docker containers, making it easily portable for hosting other sets of NPs and to be deployed on local installations.” | content_tags.findings[2] |
| finding | The October 2020 release contains 406,076 unique flat natural products and 730,441 stereochemistry-preserved records. — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | content_tags.findings[3] |
| finding | Every external molecule passes structural quality control and ChEMBL-style tautomer and ionization-state standardization. — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | content_tags.findings[4] |
| finding | Records from different sources are unified by InChIKey identity without stereochemistry while original stereochemical structures are retained when available. — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | content_tags.findings[5] |
| finding | The unification policy trades stereoisomer specificity for integration across sources of varying quality. — “The authors are well aware that different stereoisomers of a compound can have very different biological activity. The procedure described above was a necessary step to create a unified resource out of distributed databases of varying quality. Further curation will gradually improve stereochemical assignments and linkage to original source articles.” | content_tags.findings[6] |
| finding | Each unique record receives a CNP identifier plus names, cross-references, computed properties, descriptors, fingerprints, and chemical classifications. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | content_tags.findings[7] |
| finding | The database computes 73 fields for each unique natural product, while exposing a selected fraction in the web interface. — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | content_tags.findings[8] |
| finding | A five-star score summarizes the completeness and trust of names, taxonomy, literature, and source provenance. — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | content_tags.findings[9] |
| finding | All natural products receive at least one molecular name through source names, database lookup, or computed IUPAC naming. — “NPs common names in COCONUT have been retrieved, when available, from their databases of origin. The remaining NPs have been searched by InChI in major chemical databases (PubChem, ChEMBL and ChEBI) and common names and synonyms were retrieved when occurences the compound were found. Additionally to this, IUPAC names were computed for all COCONUT NPs using ChemAxon’s MolCovert [ 20 ], to add nomenclature homogeneity to the dataset. Furthemore, the IUPAC names are used as the main NP name when no official chemical name has been found nor in the original sources, nor by searching big chemical databases. Therefore, all NPs in COCONUT have at least one molecular name.” | content_tags.findings[10] |
| finding | COCONUT provides a deglycosylated representation and computed NP-likeness scores alongside other molecular features. — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | content_tags.findings[11] |
| finding | An extra provenance screen removed 24,880 SuperNatural II-only molecules that lacked positive natural-product evidence. — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | content_tags.findings[12] |
| finding | Only 31.5 percent of records have organism taxonomy, 15.4 percent have continent-level geography, and 16.6 percent have a literature reference. — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.findings[13] |
| finding | Taxonomic attribution can be mistaken when symbionts or microbiomes rather than the sampled multicellular organism synthesize a compound. — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | content_tags.findings[14] |
| finding | The database separates raw source records from unified curated records in two MongoDB collections. — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | content_tags.findings[15] |
| finding | The implementation uses React and OpenChemLib at the front end and Kotlin, Java 11, Spring, and CDK at the back end. — “The COCONUT online front-end is developed entirely with React.js [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLib library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUT back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlin and Java 11 using the Spring framework. The CDK [ 13 ] library is used to process chemical information and formats.” | content_tags.findings[16] |
| finding | The web interface, back end, and database are Dockerized for local or cloud deployment. — “COCONUT web interface, back-end and database are entirely Dockerised, allowing a quick and easy deployment on local servers and cloud. All the code, for both front-end and back-end, is available on GitHub ( https://github.com/mSorok/NaturalProductsOnline ).” | content_tags.findings[17] |
| finding | Simple search recognizes the input representation and routes it to indexed fields. — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | content_tags.findings[18] |
| finding | Substructure search combines a PubChem-fingerprint bit screen with an exact CDK pattern matcher selected by the user. — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | content_tags.findings[19] |
| finding | Similarity search uses MongoDB aggregation and an inverted index over PubChem fingerprint bits. — “Similarity search with MongoDB was implemented following the excellent ChEBML blog post on LSH-based similarity search in MongoDB [ 32 ] and adapted to Java, Kotlin and Spring data. In this approach, the MongoDB aggregation framework is used to perform inverted indexing search against PubChem fingerprints stored in a separate table and referencing COCONUT identifiers that contain the molecular features encoded by each bit.” | content_tags.findings[20] |
| finding | The full dataset and project code are freely available without restriction. — “All COCONUT data, code to process raw NPs data, data quality control and annotation, and the code for the font- and the back-end of the COCONUT online website are freely available without any restriction. The latest COCONUT data, as MongoDB full dump can be downloaded at https://coconut.naturalproducts.net/download . Code for data assembly, processing and quality control process codes is available on GitHub at https://github.com/mSorok/COCONUT . The code for the front-end and back-end is also available on GitHub at https://github.com/mSorok/NaturalProductsOnline .” | content_tags.findings[21] |
| quantity | over 120 — natural-product databases inventoried across the prior twenty years — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | content_tags.quantities[0] |
| quantity | 16% — inventoried natural-product databases no longer online — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | content_tags.quantities[1] |
| quantity | 40% — inventoried databases that are commercial — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | content_tags.quantities[2] |
| quantity | 53 — data sources contributing to COCONUT — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | content_tags.quantities[3] |
| quantity | 406076 — unique flat natural products in the October 2020 release — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | content_tags.quantities[4] |
| quantity | 730441 — records retaining stereochemistry when available — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | content_tags.quantities[5] |
| quantity | 5 to 210 — permitted heavy-atom range in structural quality control — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | content_tags.quantities[6] |
| quantity | 73 — computed fields in each unique-natural-product document — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | content_tags.quantities[7] |
| quantity | 5 stars — highest annotation quality — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | content_tags.quantities[8] |
| quantity | 1 star — lowest annotation quality — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | content_tags.quantities[9] |
| quantity | 24880 — SuperNatural II-only molecules removed by additional quality control — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | content_tags.quantities[10] |
| quantity | 31.5% — COCONUT records with at least one organism taxon — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.quantities[11] |
| quantity | 134379 — natural products with at least one organism taxon — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.quantities[12] |
| quantity | 15.4% — records with continent-level geographic provenance — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.quantities[13] |
| quantity | 66068 — natural products with known continent-level geography — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.quantities[14] |
| quantity | 16.6% — records with at least one literature reference — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.quantities[15] |
| quantity | 70730 — natural products with at least one literature reference — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.quantities[16] |
| quantity | 65% — taxonomically annotated natural products attributed only to plants — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | content_tags.quantities[17] |
| quantity | 0.5% — taxonomically annotated natural products attributed to animals — “We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUT for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.” | content_tags.quantities[18] |
| variable | data provenance (source dimension retained across aggregation) — “COCONUT data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUT release (October 2020), there are 406,076 unique “flat” (with no stereochemistry) NPs, and a total of 730,441 NPs where stereochemistry has been preserved when available.” | content_tags.variables[0] |
| variable | stereochemistry (preserved source detail but excluded from flat-record unification) — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | content_tags.variables[1] |
| variable | InChI keys without stereochemistry (identity criterion for record unification) — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | content_tags.variables[2] |
| variable | annotation level (metadata-quality outcome) — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | content_tags.variables[3] |
| variable | NP-likeness score (natural-product plausibility feature and source-specific filter) — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | content_tags.variables[4] |
| variable | producer taxonomy (biological provenance field) — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.variables[5] |
| variable | geographic location (collection or occurrence provenance field) — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | content_tags.variables[6] |
| variable | PubChem fingerprints (substructure and similarity search representation) — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | content_tags.variables[7] |
| variable | molecular descriptor values (advanced-search constraints) — “The advanced search supports searching for NPs in COCONUT according to a range of parameters, such as molecular formula, molecular descriptor values, number of rings, type of sugar moieties present in them, etc.” | content_tags.variables[8] |
| definition | natural products — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | content_tags.definitions[0] |
| definition | COCONUT — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | content_tags.definitions[1] |
| definition | annotation level — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | content_tags.definitions[4] |
| definition | deglycosylated structure representation — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | content_tags.definitions[6] |
| definition | SourceNaturalProduct — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | content_tags.definitions[7] |
| definition | UniqueNaturalProduct — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | content_tags.definitions[8] |
| definition | simple search — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | content_tags.definitions[9] |
| definition | substructure search — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | content_tags.definitions[10] |
| definition | REST API — “A REST API has been developed for COCONUT online in order to permit programmatic querying and facilitate its integration in workflows. It relies on Kotlin API functionalities and it’s usage, together with some examples, is described in detail in the documentation section of the website ( https://coconut.naturalproducts.net/documentation ).” | content_tags.definitions[11] |
Concepts
terms this paper carries, each with the sentence it was cut from — the layer that joins one paper to another. Terms with no quote yet are on CONCEPTS.html.
| concept | Natural products — small molecules produced by living organisms — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[0] |
| concept | small molecules — not defined in this paper's own words; molecular scale of natural products — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[1] |
| concept | living organisms — not defined in this paper's own words; biological producers defining natural products — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[2] |
| concept | bioactive — not defined in this paper's own words; potential biological activity motivating applications — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[3] |
| concept | COCONUT — not defined in this paper's own words; aggregated open natural-product collection and online database — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[4] |
| concept | aggregated dataset — not defined in this paper's own words; collection assembled across open sources — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[5] |
| concept | elucidated and predicted NPs — not defined in this paper's own words; two provenance statuses included in the collection — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[6] |
| concept | web interface — not defined in this paper's own words; interactive access layer for browsing and retrieval — “Natural products (NPs) are small molecules produced by living organisms with potential applications in pharmacology and other industries as many of them are bioactive. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTs (COCONUT): an aggregated dataset of elucidated and predicted NPs collected from open sources and a web interface to browse, search and easily and quickly download NPs. COCONUT web is freely available at https://coconut.naturalproducts.net .” | concepts[7] |
| concept | generalistic NPs database — not defined in this paper's own words; cross-domain alternative to specialized collections — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | concepts[8] |
| concept | drug discovery — not defined in this paper's own words; application area for natural-product fragments — “Natural products (NPs) have received constant attention from the scientific community due to their relevance in drug discovery, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural products databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs database, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTs (COCONUT) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUT have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discovery and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural products database, maintained at https://coconut.naturalproducts.net .” | concepts[9] |
| concept | REST API — not defined in this paper's own words; programmatic query interface — “The COCONUT database is free and open to all users and there is no login required to access it. Its web interface allows diverse simple searches (e.g. by molecule name, InChI, InChI key, SMILES, drawn structure, molecular formula), advanced search by molecular features, together with substructure and similarity searches. Users can also download the whole dataset or search results in different formats. The database can be queried programmatically via a REST API, which facilitates COCONUT integration in workflows. The web interface, the back-end and the database are deployed as Docker containers, making it easily portable for hosting other sets of NPs and to be deployed on local installations.” | concepts[10] |
| concept | Docker containers — not defined in this paper's own words; portable deployment units — “The COCONUT database is free and open to all users and there is no login required to access it. Its web interface allows diverse simple searches (e.g. by molecule name, InChI, InChI key, SMILES, drawn structure, molecular formula), advanced search by molecular features, together with substructure and similarity searches. Users can also download the whole dataset or search results in different formats. The database can be queried programmatically via a REST API, which facilitates COCONUT integration in workflows. The web interface, the back-end and the database are deployed as Docker containers, making it easily portable for hosting other sets of NPs and to be deployed on local installations.” | concepts[11] |
| concept | chemical data sources — not defined in this paper's own words; upstream collections used for COCONUT — “COCONUT is assembled from a large number of chemical data sources (Table 1 ), from which NPs have been thoroughly extracted, curated, processed and annotated. The resulting NPs collection is presented within the full-fledged chemical database developed specially for this purpose (Fig. 1 ). Data curation and processing steps, together with the construction of the web interface and the description of available features are described below.” | concepts[12] |
| concept | quality control — not defined in this paper's own words; structural acceptance screening — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | concepts[13] |
| concept | registration procedure — not defined in this paper's own words; entry procedure following structure checks — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | concepts[14] |
| concept | heavy atoms — not defined in this paper's own words; structure-size basis for quality control — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | concepts[15] |
| concept | biggest connected structure — not defined in this paper's own words; component retained from a submitted record — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | concepts[16] |
| concept | tautomers — not defined in this paper's own words; chemical states standardized during curation — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | concepts[17] |
| concept | ionisation states — not defined in this paper's own words; charge states standardized during curation — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | concepts[18] |
| concept | ChEMBL chemical structure curation pipeline — not defined in this paper's own words; source of chemical standardization rules — “Every molecule collected from external sources passed a quality control and a registration procedure, where its structure is checked for size (between 5 and 210 heavy atoms), connectivity (only the biggest connected structure is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomers and ionisation states are standardized following the ChEMBL chemical structure curation pipeline [ 12 ].” | concepts[19] |
| concept | InChI keys without stereochemistry — not defined in this paper's own words; flat-record identity key — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | concepts[20] |
| concept | stereochemistry — not defined in this paper's own words; source detail excluded from unification but retained when available — “Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistry. This unification step is performed without stereochemistry, as in different data sources stereochemistry is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistry is preserved and can be accessed for each NP entry.” | concepts[21] |
| concept | stereoisomers — not defined in this paper's own words; variants that can differ in biological activity — “The authors are well aware that different stereoisomers of a compound can have very different biological activity. The procedure described above was a necessary step to create a unified resource out of distributed databases of varying quality. Further curation will gradually improve stereochemical assignments and linkage to original source articles.” | concepts[22] |
| concept | unique identifier — not defined in this paper's own words; CNP-prefixed record identifier — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[23] |
| concept | molecular properties — not defined in this paper's own words; computed chemical record features — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[24] |
| concept | descriptors — not defined in this paper's own words; computed molecular attributes — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[25] |
| concept | fingerprints — not defined in this paper's own words; computed structural bit representations — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[26] |
| concept | MongoDB — not defined in this paper's own words; document-oriented database storing source and unified records — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[27] |
| concept | ClassyFire — not defined in this paper's own words; hierarchical chemical classifier — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[28] |
| concept | hierarchical chemical classification — not defined in this paper's own words; nested chemical-class assignment — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[29] |
| concept | Murcko frameworks — not defined in this paper's own words; computed structural frameworks — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[30] |
| concept | Ertl Functional Groups — not defined in this paper's own words; computed functional-group representation — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[31] |
| concept | deepSMILES — not defined in this paper's own words; SMILES adaptation for deep-learning use — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[32] |
| concept | adaptation of SMILES — not defined in this paper's own words; representation class of deepSMILES — “Each unique NP is then assigned a unique identifier, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular properties, descriptors and fingerprints (full list in Table 2 ) are computed using the in-build CDK [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUT web interface. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature references (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDB. The chemical classification of all NPs in COCONUT is performed with ClassyFire [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFire provides a hierarchical chemical classification of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworks [ 15 ], Ertl Functional Groups [ 16 ] and deepSMILES [ 17 ]. DeepSMILES is an adaptation of SMILES for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.” | concepts[33] |
| concept | annotation level — not defined in this paper's own words; five-star metadata-quality summary — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | concepts[34] |
| concept | 5-star-based system — not defined in this paper's own words; scale used for annotation quality — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | concepts[35] |
| concept | trusted data provenances — not defined in this paper's own words; enumerated sources contributing to higher annotation scores — “Last, the annotation level of each NP in COCONUT is computed. It is a 5-star-based system, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature reference and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenances. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUT NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation level is 3. The annotation level is represented with stars on the NP page.” | concepts[36] |
| concept | IUPAC names — not defined in this paper's own words; computed systematic names — “NPs common names in COCONUT have been retrieved, when available, from their databases of origin. The remaining NPs have been searched by InChI in major chemical databases (PubChem, ChEMBL and ChEBI) and common names and synonyms were retrieved when occurences the compound were found. Additionally to this, IUPAC names were computed for all COCONUT NPs using ChemAxon’s MolCovert [ 20 ], to add nomenclature homogeneity to the dataset. Furthemore, the IUPAC names are used as the main NP name when no official chemical name has been found nor in the original sources, nor by searching big chemical databases. Therefore, all NPs in COCONUT have at least one molecular name.” | concepts[37] |
| concept | Sugar Removal Utility — not defined in this paper's own words; sugar detection and removal software — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | concepts[38] |
| concept | deglycosylated structure representation — not defined in this paper's own words; representation supporting aglycon analysis — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | concepts[39] |
| concept | NP-likeness score — not defined in this paper's own words; natural-product plausibility score — “Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUT. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUT have been analysed for sugar moieties presence, and a deglycosylated structure representation was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utility [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness score [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.” | concepts[40] |
| concept | NPclassifier — not defined in this paper's own words; neural structural classifier specialized for natural products — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | concepts[41] |
| concept | deep neural network-based structural classification tool — not defined in this paper's own words; tool class of NPclassifier — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | concepts[42] |
| concept | genuine or predicted NP — not defined in this paper's own words; admissible natural-product status for uncertain source records — “At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUT, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUT and be considered as a genuine or predicted NP, such a molecule has to have a strictly positive NP-likeness score, be classified as a NP by NPclassifier [ 24 ], a deep neural network-based structural classification tool specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality control have been removed from COCONUT, until further proof of their natural provenance.” | concepts[43] |
| concept | literature reference — not defined in this paper's own words; publication provenance for a natural product — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | concepts[44] |
| concept | NCBI taxonomy identifier — not defined in this paper's own words; standard organism identifier — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | concepts[45] |
| concept | geographic location — not defined in this paper's own words; location provenance for producer occurrence or collection — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | concepts[46] |
| concept | manual curation — not defined in this paper's own words; human recovery of missing provenance links — “The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organisms. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature reference, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifier and the geographic location where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUT was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUT are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic location (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature reference. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curation, but the amount of data in COCONUT is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.” | concepts[47] |
| concept | document-oriented NoSQL database program — not defined in this paper's own words; database class of MongoDB — “All COCONUT data is stored with MongoDB, a cross-platform document-oriented NoSQL database program. The smallest unit in MongoDB is a document, composed of key and value pairs that are similar to JSON objects. Documents of the same nature are organized in collections, which are the equivalent of the SQL-based databases tables. MongoDB is particularly adapted to big and complex data, supports multiple indexing, including text indexing allowing enhanced text search in text-indexed fields and contains a wide range of in-build search and analysis functions.” | concepts[48] |
| concept | SourceNaturalProduct — not defined in this paper's own words; collection of original source records — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | concepts[49] |
| concept | UniqueNaturalProduct — not defined in this paper's own words; collection of unified curated records — “Two major collections are present in the COCONUT database: SourceNaturalProduct, which contains the original NPs data collected from the open sources, and UniqueNaturalProduct, the unified and curated collection of NPs. The full version of COCONUT with all the calculated features can be accessed as a MongoDB dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interface and making them searchable through the advanced search interface are welcome via the COCONUT GitHub tracker (see below).” | concepts[50] |
| concept | React.js — not defined in this paper's own words; front-end user-interface library — “The COCONUT online front-end is developed entirely with React.js [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLib library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUT back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlin and Java 11 using the Spring framework. The CDK [ 13 ] library is used to process chemical information and formats.” | concepts[51] |
| concept | OpenChemLib — not defined in this paper's own words; chemical editor used for search — “The COCONUT online front-end is developed entirely with React.js [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLib library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUT back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlin and Java 11 using the Spring framework. The CDK [ 13 ] library is used to process chemical information and formats.” | concepts[52] |
| concept | Kotlin — not defined in this paper's own words; back-end implementation language — “The COCONUT online front-end is developed entirely with React.js [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLib library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUT back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlin and Java 11 using the Spring framework. The CDK [ 13 ] library is used to process chemical information and formats.” | concepts[53] |
| concept | Java 11 — not defined in this paper's own words; back-end implementation platform — “The COCONUT online front-end is developed entirely with React.js [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLib library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUT back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlin and Java 11 using the Spring framework. The CDK [ 13 ] library is used to process chemical information and formats.” | concepts[54] |
| concept | Spring framework — not defined in this paper's own words; back-end application framework — “The COCONUT online front-end is developed entirely with React.js [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLib library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUT back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlin and Java 11 using the Spring framework. The CDK [ 13 ] library is used to process chemical information and formats.” | concepts[55] |
| concept | CDK — not defined in this paper's own words; chemical information processing library — “The COCONUT online front-end is developed entirely with React.js [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLib library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUT back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlin and Java 11 using the Spring framework. The CDK [ 13 ] library is used to process chemical information and formats.” | concepts[56] |
| concept | simple searches — not defined in this paper's own words; multi-identifier header queries — “The COCONUT database is free and open to all users and there is no login required to access it. Its web interface allows diverse simple searches (e.g. by molecule name, InChI, InChI key, SMILES, drawn structure, molecular formula), advanced search by molecular features, together with substructure and similarity searches. Users can also download the whole dataset or search results in different formats. The database can be queried programmatically via a REST API, which facilitates COCONUT integration in workflows. The web interface, the back-end and the database are deployed as Docker containers, making it easily portable for hosting other sets of NPs and to be deployed on local installations.” | concepts[57] |
| concept | text indexing — not defined in this paper's own words; MongoDB name and synonym retrieval mechanism — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | concepts[58] |
| concept | regular expressions — not defined in this paper's own words; input-type recognition mechanism — “The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUT ids and molecular formulas. Name search uses native MongoDB text indexing, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressions, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.” | concepts[59] |
| concept | PubChem fingerprints — not defined in this paper's own words; bit representation used for chemical retrieval — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | concepts[60] |
| concept | $allBitsSet function — not defined in this paper's own words; MongoDB fingerprint-containment filter — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | concepts[61] |
| concept | Ullmann — not defined in this paper's own words; default exact substructure matcher — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | concepts[62] |
| concept | Vento-Foggia — not defined in this paper's own words; alternative exact substructure matcher — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | concepts[63] |
| concept | depth-first (DF) — not defined in this paper's own words; alternative exact substructure matcher — “Searching for an exact substructure in a MongoDB database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprints of choice (in COCONUT are used the PubChem fingerprints) to be precomputed and stored as a list of bytes (BinData type in MongoDB). The query molecule (substructure) then needs to have its fingerprint to be also computed and to be matched against the database using the $allBitsSet function [ 28 ]. This native to MongoDB function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmann [ 29 ], the Vento-Foggia [ 30 ] and the depth-first (DF) [ 31 ] pattern matching approaches, all performed using the CDK in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmann, method is to be privileged by users unfamiliar with the intricacies of such matching.” | concepts[64] |
| concept | LSH-based similarity search — not defined in this paper's own words; fingerprint similarity retrieval approach — “Similarity search with MongoDB was implemented following the excellent ChEBML blog post on LSH-based similarity search in MongoDB [ 32 ] and adapted to Java, Kotlin and Spring data. In this approach, the MongoDB aggregation framework is used to perform inverted indexing search against PubChem fingerprints stored in a separate table and referencing COCONUT identifiers that contain the molecular features encoded by each bit.” | concepts[65] |
| concept | advanced search — not defined in this paper's own words; feature-constrained natural-product retrieval — “The advanced search supports searching for NPs in COCONUT according to a range of parameters, such as molecular formula, molecular descriptor values, number of rings, type of sugar moieties present in them, etc.” | concepts[66] |
Full text — tagged
COCONUTterm online: Collection of Open Natural Productsterm database Abstract Natural productsterm (NPs) are small moleculesterm produced by living organismsterm with potential applications in pharmacology and other industries as many of them are bioactiveterm. This potential raised great interest in NP research around the world and in different application fields, therefore, over the years a multiplication of generalistic and thematic NP databases has been observed. However, there is, at this moment, no online resource regrouping all known NPs in just one place, which would greatly simplify NPs research and allow computational screening and other in silico applications. In this manuscript we present the online version of the COlleCtion of Open Natural prodUcTsterm (COCONUTterm): an aggregated datasetterm of elucidated and predicted NPsterm collected from open sources and a web interfaceterm to browse, search and easily and quickly download NPs. COCONUTterm web is freely available at https://coconutterm.naturalproducts.net .
Keywords: Natural productsterm, Database, NoSQL, MongoDBterm, Molecular similarity search, Molecular substructure searchterm·def Introduction Natural productsterm (NPs) have received constant attention from the scientific community due to their relevance in drug discoveryterm, chemical ecology and molecular biology in general. In a recently published review on NPs databases [ 1 ] we inventoried over 120 natural productsterm databases that have been published and used in the last 20 years. However, 16% of these are not available online anymore, 40% are commercial and their content cannot be easily accessed. The open resources are generally either specialized on a particular type of NPs, either lack annotations. For instance, the catalog of NPs from the ZINC database [ 2 ] is composed of over 80,000 entries, some of which can be purchased, but apart from their structure and that they are from natural origin, no additional information is provided. Super Natural II [ 3 ] is considered as the largest among all the NP databases, is accessible online in 2020, but it seems not to be maintained anymore and is mainly composed of compounds that can be purchased. Another recent database, NPAtlas [ 4 ], is constantly growing and extremely well annotated, but it is focusing on microbial NPs only. Another major NPs category, plant-produced compounds, also called phytochemicals, is available in several popular and well maintained databases, such as NuBBEDB [ 5 ], KnapSack [ 6 ], CMAUP [ 7 ] and TCM@Taiwan [ 8 ]. In addition to these relatively big databases, there is a plethora of smaller, more specialized NPs collections, such as FooDB [ 9 ], a user-friendly database hosting a relatively large number of NPs that are found in food. There is, therefore, a need for a generalistic NPs databaseterm, that will efficiently aggregate NPs information from various sources, improve its annotation and offer a pleasant user experience. With this ultimate goal in mind, we first assembled the most complete up-to-date COlleCtion of Open Natural ProdUcTsterm (COCONUTterm) that we have been continuously curating and annotating. Studies [ 10 , 11 ] showed that fragments from NPs present in COCONUTterm have high diversity and structural complexity, which makes it, among other possible applications, a suitable source for drug discoveryterm and can be included in drug design pipelines. Our next step was to make this data available to the scientific community as a full-fledged online natural productsterm database, maintained at https://coconutterm.naturalproducts.net .
The COCONUTterm database is free and open to all users and there is no login required to access it. Its web interfaceterm allows diverse simple searchesterm (e.g. by molecule name, InChI, InChI key, SMILES, drawn structure, molecular formula), advanced searchterm by molecular features, together with substructure and similarity searches. Users can also download the whole dataset or search results in different formats. The database can be queried programmatically via a REST APIterm, which facilitates COCONUTterm integration in workflows. The web interfaceterm, the back-end and the database are deployed as Docker containersterm, making it easily portable for hosting other sets of NPs and to be deployed on local installations.
Construction and content COCONUTterm is assembled from a large number of chemical data sourcesterm (Table 1 ), from which NPs have been thoroughly extracted, curated, processed and annotated. The resulting NPs collection is presented within the full-fledged chemical database developed specially for this purpose (Fig. 1 ). Data curation and processing steps, together with the construction of the web interfaceterm and the description of available features are described below.
Public databases and datasets from which COCONUTterm was assembled Construction and curation of COCONUTterm online Data provenanceterm·var, model and content COCONUTterm data has been extracted from 53 various data sources and several manually collected from literature sets, as shown in Table 1 . In the current COCONUTterm release (October 2020), there are 406,076 unique “flat” (with no stereochemistryterm) NPs, and a total of 730,441 NPs where stereochemistryterm has been preserved when available.
Every molecule collected from external sources passed a quality controlterm and a registration procedureterm, where its structure is checked for size (between 5 and 210 heavy atomsterm), connectivity (only the biggest connected structureterm is kept), presence of pseudo-atoms, if implicit and explicit hydrogens are correct, and if the bonds are correct and the valences are conserved. The Kekulé representation is also assigned to the aromatic systems of each compound. At this step, tautomersterm and ionisation statesterm are standardized following the ChEMBL chemical structure curation pipelineterm [ 12 ].
Then, NPs from different provenance are unified based on the identity of their InChI keys without stereochemistryterm. This unification step is performed without stereochemistryterm, as in different data sources stereochemistryterm is not systematically present and can also be represented differently. When available, the original molecular structure with stereochemistryterm is preserved and can be accessed for each NP entry.
The authors are well aware that different stereoisomersterm of a compound can have very different biological activity. The procedure described above was a necessary step to create a unified resource out of distributed databases of varying quality. Further curation will gradually improve stereochemical assignments and linkage to original source articles.
Each unique NP is then assigned a unique identifierterm, composed of the “CNP” prefix and 7 digits. An automatic curation for NP metadata is performed, comprising the retrieval of its official name, synonyms, cross-references to other major chemical databases. Then, a range of molecular propertiesterm, descriptorsterm and fingerprintsterm (full list in Table 2 ) are computed using the in-build CDKterm [ 13 ] libraries. As the number of the computed properties is quite big (73 fields in each document corresponding to one unique NP), only a selected fraction of them is displayed on the COCONUTterm web interfaceterm. Finally, the first round of automatic curation of NP metadata, in particular the molecular name synonyms, cross-references with other major chemical databases, correction of the literature referencesterm (PubMed identifiers and DOIs) and taxonomy is performed. All original data, unified NPs and the derived and calculated information are stored in MongoDBterm. The chemical classification of all NPs in COCONUTterm is performed with ClassyFireterm [ 14 ], and, when available, is displayed in the corresponding section of the compound page. ClassyFireterm provides a hierarchical chemical classificationterm of chemical compounds and enables grouping NPs by their chemical class. Additionally, frameworks facilitating NPs analyses for their chemical and therapeutic properties are computed for NPs, such as Murcko frameworksterm [ 15 ], Ertl Functional Groupsterm [ 16 ] and deepSMILESterm [ 17 ]. DeepSMILESterm is an adaptation of SMILESterm for use in deep machine learning of chemical structures. Due to the increased usage of deep learning in chemistry, it is indeed interesting to provide this new chemical representation type pre-computed for NPs.
Molecular features present in COCONUTterm and in their disponibility in the web interfaceterm Last, the annotation levelterm of each NP in COCONUTterm is computed. It is a 5-star-based systemterm, where 1 star is the lowest annotation quality (no verified common name, no taxonomic provenance annotation, no literature referenceterm and no trusted data source) and 5 stars is the highest quality, with all the intermediate annotation qualities reflected by 2, 3 and 4 stars. Only ChEBI [ 18 ], KNApSAcK [ 6 ], ChEMBL [ 19 ], CMAUP [ 7 ], NPAtlas [ 4 ] and, of course, the manually picked data are considered as trusted data provenancesterm. For example, caryolivine (CNP0235854) has a 5-stars annotation because it has a verified common name, is known to be produced by Caryomene olivascens , a plant, is associated to a scientific publication and is present in KNApSAcK. The COCONUTterm NP CNP0330764 has no verified common name, only a computed IUPAC one, but is present in CMAUP and is known to be produced by a range of plants, therefore it’s annotation levelterm is 3. The annotation levelterm is represented with stars on the NP page.
Natural productterm naming NPs common names in COCONUTterm have been retrieved, when available, from their databases of origin. The remaining NPs have been searched by InChI in major chemical databases (PubChem, ChEMBL and ChEBI) and common names and synonyms were retrieved when occurences the compound were found. Additionally to this, IUPAC namesterm were computed for all COCONUTterm NPs using ChemAxon’s MolCovert [ 20 ], to add nomenclature homogeneity to the dataset. Furthemore, the IUPAC namesterm are used as the main NP name when no official chemical name has been found nor in the original sources, nor by searching big chemical databases. Therefore, all NPs in COCONUTterm have at least one molecular name.
Computed molecular features Figure 2 demonstrates the distributions and relationships of a small selection of computed molecular features within COCONUTterm. Sugar moieties occur frequently in NPs and have an important impact on their bioactivities and physicochemical properties. However, they are often redundant and therefore obstruct the study of the aglycon. For this reason, NPs in COCONUTterm have been analysed for sugar moieties presence, and a deglycosylated structure representationterm was made available in the database. Sugar moieties manipulations were performed using the Sugar Removal Utilityterm [ 21 ]. To track their influence on other features, their absence and presence are colour-mapped (no sugar moiety in the molecular structure in blue, and the presence of at least one sugar moiety in orange). The wide molecular weight range is typical for NPs; it is, however, interesting to notice its correlation with the number of oxygen atoms in the molecule, regardless of the presence and absence of sugar. Another interesting correlation to be noted is between the molecular weight and the nitrogen atom number in sugar-free molecules. The NP-likeness scoreterm [ 22 ], trained on high-quality NPs dataset and computed with NaPLeS [ 23 ], which was trained on high-quality NPs dataset, has a typical distribution for an NPs set, where most molecules have a positive score.
Pair plot of a selection of five of the molecular features available in COCONUTterm. Colour mapping corresponds to the presence (yellow) and absence (blue) of glycosidic moieties in the molecular structures of the NPs in COCONUTterm At this point, an additional NPs curation step has been performed, due to the possible inconsistency in genuine NPs of one of the used sources, SuperNatural II. NPs that are not occuring in other datasets used to assemble COCONUTterm, but only in SuperNatural II, have been thoroughly tested. To be kept in COCONUTterm and be considered as a genuine or predicted NPterm, such a molecule has to have a strictly positive NP-likeness scoreterm, be classified as a NP by NPclassifierterm [ 24 ], a deep neural network-based structural classification toolterm specialised in NPs or have a sugar moiety in its structure. The 24,880 molecules from SuperNatural II that didn’t pass this additional quality controlterm have been removed from COCONUTterm, until further proof of their natural provenance.
Counting rings in a molecule can become a complex task, as the outer perimeter of two fused rings can be counted as one big ring. With more condensed rings, the number of fused ring perimeters (aka as the set of all rings) can grow steeply. In Fig. 2 , only the minimal ring count (the minimal cycle base) is represented.
Natural productterm annotation The particularity of NPs, opposed to synthetic molecules, which constitute the biggest part of chemical databases, resides in their production by living organismsterm. Therefore, in addition to their structure and computable structural properties, NPs need to be annotated with at least one literature referenceterm, mentioning where, when and from which organism the NP was isolated. As a direct consequence, an NP entry should be associated with at least one organism, preferentially with an NCBI taxonomy identifierterm and the geographic locationterm where the organism is naturally occuring or has been collected. Regrettably, this metadata is often omitted in public databases from which COCONUTterm was assembled. Therefore, only 31.5% (134,379) of NPs in COCONUTterm are annotated with at least one organism taxa, for 15.4% (66,068) of NPs the geographic locationterm (on the continent level) of the organism occurence or collection is known and only 16.6% (70,730) of NPs have at least one literature referenceterm. These numbers combine both the retrieval of the original NP annotations from their sources and our efforts to retrieve more extensive information from major trusted chemical databases, PubChem [ 25 ], ChEMBL [ 19 ], ChEBI [ 18 ], CMAUP [ 7 ] and KnapSacK [ 6 ]. Despite our efforts, most of the links between the original publication of the structure elucidation of an NP and its reference, source organism and its geographical location are still missing. A possible solution to fill these gaps is manual curationterm, but the amount of data in COCONUTterm is redhibitory for even considering this approach. Another solution is to use unsupervised machine learning for optical recognition approaches, to parse modern peer-reviewed literature and books to re-establish links between NPs structures and their provenance.
We analysed the taxonomic classification of known NPs producers together with overlaps in NPs production between superkingdom for the 31% of the NPs in COCONUTterm for which the provenance organism is known (Fig. 3 ). Here are distinguished five taxonomic categories: plants, bacteria, fungi, animals and marine. The last one is not a proper monoclade classification, but rather reflects a group of organisms that are found only in marine and oceanic environments, and therefore can overlap in terms of its species and NP content with other categories, which are more stringent taxonomically. A large part (65%) of these annotated NPs are produced only by plants, and only very few (0.5%) are from animal origin. Main overlaps in terms of NP production between the taxonomic kingdoms are between plants and marine organisms (which is unsurprising, as there can be real plants among the marine entities) and surprisingly between plants and fungi. The other overlaps between taxonomic kingdoms are not as significant. It needs to be pointed out here that multicellular organisms, such as plants, animals and some of the fungi are most of the time in symbiosis with microorganisms, in particular bacteria. Therefore, NPs isolated from such a multicellular organism can be synthesized and secreted by their symbionts or microbiomes, and therefore mistakenly assigned to an incorrect organism.
Overlap of NPs taxonomic provenance in COCONUTterm The geographic locationterm of the collection or the natural presence of the NP-producing organism is a piece of information that is even more difficult to obtain. Nowadays, a range of organisms, and in particular plants, can be found in different parts of the planet due to globalisation and their success in human consumption (e.g. garlic, tomatoes, curcuma or ginger). It is, therefore, difficult, if not impossible, to determine their original provenance. Also, the geographical information is often omitted in literature and most NPs databases. When available, the geographical provenance is stored in the MongoDBterm dump of COCONUTterm, but not displayed on the website.
For NPs where geographical information is available, it appears that most of them are produced by organisms that have been isolated in Asia (Fig. 4 ). This bias is introduced by the intensive study by scientists of the traditional Chinese and Indian medicines and by the big efforts in isolation and elucidation of NPs from medicinal plants. NPs from the African continent are also well represented in COCONUTterm (Fig. 4 ), mainly due to the scientific interest in African traditional medicines and African biodiversity. There is, for now, no data from the biodiversity of the Australian continent, and only very little data for NPs isolated from endemic European organisms. NPs from the Americas are mainly extracted and solved while Brazilian and Mexican biodiversity exploration. Only a few NPs are present in more than one continent, mainly in Asia and Africa, and the overlap values are biased by the very different NPs set sizes between the different continents.
Overlap of NPs geographic provenance in COCONUTterm Web interfaceterm and technical specificities All COCONUTterm data is stored with MongoDBterm, a cross-platform document-oriented NoSQL database programterm. The smallest unit in MongoDBterm is a document, composed of key and value pairs that are similar to JSON objects. Documents of the same nature are organized in collections, which are the equivalent of the SQL-based databases tables. MongoDBterm is particularly adapted to big and complex data, supports multiple indexing, including text indexingterm allowing enhanced text search in text-indexed fields and contains a wide range of in-build search and analysis functions.
Two major collections are present in the COCONUTterm database: SourceNaturalProductterm, which contains the original NPs data collected from the open sources, and UniqueNaturalProductterm, the unified and curated collection of NPs. The full version of COCONUTterm with all the calculated features can be accessed as a MongoDBterm dump in the Downloads section of the website. Requests for displaying additional crucial features in the web interfaceterm and making them searchable through the advanced searchterm interface are welcome via the COCONUTterm GitHub tracker (see below).
The COCONUTterm online front-end is developed entirely with React.jsterm [ 26 ], a JavaScript library to build responsive and efficient user interfaces. The OpenChemLibterm library [ 27 ] is used to handle the chemical editor for the search functions. The COCONUTterm back-end, allowing to process the front-end requests and to communicate with the database is written in Kotlinterm and Java 11term using the Spring frameworkterm. The CDKterm [ 13 ] library is used to process chemical information and formats.
COCONUTterm web interfaceterm, back-end and database are entirely Dockerised, allowing a quick and easy deployment on local servers and cloud. All the code, for both front-end and back-end, is available on GitHub ( https://github.com/mSorok/NaturalProductsOnline ).
Searching the database COCONUTterm online has been developed to be a full-fledged chemical database and in particular to fit the NPs structural and annotational particularities, with all the subsequent functions. At the moment, the chemical search is uncommon with MongoDBterm, therefore several approaches have been implemented to run molecular substructure and similarity searches.
Simple searchterm·def The so-called “simple” search can be performed using the header search bar. The users can enter there molecule names (e.g. “curcumin”), SMILES, InChI, InChi key, COCONUTterm ids and molecular formulas. Name search uses native MongoDBterm text indexingterm, allowing fuzzy, flexible search in the “name” and “synonyms” fields. The input string type is first identified using regular expressionsterm, then the DB is queried against the appropriate fields, and the result, when exists, is returned to the front-end.
Substructure searchterm·def implementation Searching for an exact substructure in a MongoDBterm database of molecules appears to be surprisingly easy. Each molecule in the database needs to have their fingerprintsterm of choice (in COCONUTterm are used the PubChem fingerprintsterm) to be precomputed and stored as a list of bytes (BinData type in MongoDBterm). The query molecule (substructure) then needs to have its fingerprintterm to be also computed and to be matched against the database using the $allBitsSet functionterm [ 28 ]. This native to MongoDBterm function allows to select documents in a collection where a BinData field has all the query bits set to “on” (but can have bits set to “on” that are not present in the query). To confirm the substructure match, the user can select between the default Ullmannterm [ 29 ], the Vento-Foggiaterm [ 30 ] and the depth-first (DF)term [ 31 ] pattern matching approaches, all performed using the CDKterm in-build algorithms. These three pattern matching techniques tend to, generally, return very similar results, the difference between them lying rather in their approach to matching substructures, therefore the usage of the default, Ullmannterm, method is to be privileged by users unfamiliar with the intricacies of such matching.
Similarity search implementation Similarity search with MongoDBterm was implemented following the excellent ChEBML blog post on LSH-based similarity searchterm in MongoDBterm [ 32 ] and adapted to Java, Kotlinterm and Spring data. In this approach, the MongoDBterm aggregation framework is used to perform inverted indexing search against PubChem fingerprintsterm stored in a separate table and referencing COCONUTterm identifiers that contain the molecular features encoded by each bit.
Advanced searchterm The advanced searchterm supports searching for NPs in COCONUTterm according to a range of parameters, such as molecular formula, molecular descriptor valuesterm·var, number of rings, type of sugar moieties present in them, etc.
Querying COCONUTterm through the API A REST APIterm has been developed for COCONUTterm online in order to permit programmatic querying and facilitate its integration in workflows. It relies on Kotlinterm API functionalities and it’s usage, together with some examples, is described in detail in the documentation section of the website ( https://coconutterm.naturalproducts.net/documentation ).
Documentation Complete documentation describing COCONUTterm, its data and functionalities are available at the documentation section of the website https://coconutterm.naturalproducts.net/documentation .
Utility and discussion The online COCONUTterm database is an open tool for researchers in the natural productsterm community. COCONUTterm is the biggest collection of NPs in 2020 and the data it contains already benefits researchers in NPs with various aims, such as biodiversity research and drug discoveryterm. The web interfaceterm allows querying and parsing the data collection in various, chemically relevant ways with adequate performance. It is also the first big chemical database using MongoDBterm as a storage management system.
A wide range of molecular descriptorsterm are pre-computed and literature, producer taxonomyterm·var and their geography are as much annotated as currently possible without extensive manual curationterm. The web database can be searched in multiple ways, by molecular structure, by compound name and by molecular features, making this repository a complete chemical database. The user interface is modern and easy to use. Besides, the whole content of COCONUTterm is available for download in multiple formats.
In the close future, COCONUTterm will support user registration to enable user-driven NPs curation and submission and will undergo a better data annotation, in particular regarding the organisms that are producing the NPs, their geography and the corresponding literature, using deep learning approaches.
Feedback Bugs, annotation issues and requests of new COCONUTterm entries or re-annotation of existing ones can be reported at the project issues tracker ( https://github.com/mSorok/NaturalProductsOnline/issues ). Suggestions for new features are also welcome.
Availability All COCONUTterm data, code to process raw NPs data, data quality controlterm and annotation, and the code for the font- and the back-end of the COCONUTterm online website are freely available without any restriction. The latest COCONUTterm data, as MongoDBterm full dump can be downloaded at https://coconutterm.naturalproducts.net/download . Code for data assembly, processing and quality controlterm process codes is available on GitHub at https://github.com/mSorok/COCONUTterm . The code for the front-end and back-end is also available on GitHub at https://github.com/mSorok/NaturalProductsOnline .
Conclusions COCONUTterm is the largest open collection of elucidated and predicted NPsterm at this time. It has a great potential of being of particular importance for the NPs research community as it gathers most of open NPs knowledge in one single place, and makes it easily accessible and queryable.
The final aim of COCONUTterm is to provide to the scientific community NPs structures and their provenance, i.e.. organisms that synthesize them and geographic locationterm of the latter. However, a lot of data curation, in particular using new generation deep learning-based methods of extracting information from publications and books, together with website functionalities developments are still need to be done for COCONUTterm, but the database as it is now is already an important tool to facilitate NPs and medicinal chemistry research.
Acknowledgements Not applicable.
Abbreviations Natural productsterm COlleCtion of Open Natural ProdUcTsterm Chemistry Development Kit International Union of Pure and Applied Chemistry Application Programming Interface REpresentational State Transfer Digital Object Identifier JavaScript Object Notation (non-/not only) Structured Query Language Authors’ contributions MS coordinated the study, participated in the implementation of the front-end and the backend, collected; processed and curated the COCONUTterm data and wrote the manuscript. PM participated in the implementation of the front-end and the back-end of COCONUTterm online. KR and MAY helped with COCONUTterm data curation. KR designed the COCONUTterm logo. CS designed and supervised the study. All authors read and approved the final manuscript.
Funding Open Access funding enabled and organized by Projekt DEAL. This work was supported by the German Research Foundation within the framework CRC1127 ChemBioSys.
Availability of data and materials the source code of the web interfaceterm and the back-end is available on GitHub at https://github.com/mSorok/NaturalProductsOnline . The data was curated and processed using the COCONUTterm code suite available on GitHub at https://github.com/mSorok/COCONUTterm . All COCONUTterm data can be accessed on the website at https://coconutterm.naturalproducts.net/ and downloaded entirely or partially in several formats (MongoDBterm dump, SDF and SMI (SMILES)).
Competing interests The authors declare no competing interests.
Footnotes Contributor Information Maria SorokinapersName, Email: maria.sorokinapersName@uni-jena.de.
Peter MerseburgerpersName, Email: peter.merseburgerpersName@uni-jena.de.
Kohulan RajanpersName, Email: kohulan.rajanpersName@uni-jena.de.
Mehmet Aziz YirikpersName, Email: yirikpersName.mehmetaziz@uni-jena.de.
Christoph SteinbeckpersName, Email: christoph.steinbeckpersName@uni-jena.de.
References Associated Data Data Availability Statement the source code of the web interfaceterm and the back-end is available on GitHub at https://github.com/mSorok/NaturalProductsOnline . The data was curated and processed using the COCONUTterm code suite available on GitHub at https://github.com/mSorok/COCONUTterm . All COCONUTterm data can be accessed on the website at https://coconutterm.naturalproducts.net/ and downloaded entirely or partially in several formats (MongoDBterm dump, SDF and SMI (SMILES)).