Carlos P CantalapiedrapersName, Ana Hernández-PlazapersName, Ivica LetunicpersName, Peer BorkpersName, Jaime Huerta-CepaspersName (2021date). “eggNOG-mapper v2: Functional Annotation, Orthology Assignments, and Domain Prediction at the Metagenomic Scale.” Molecular Biology and Evolutiontitle·j.
Object of study — the coverage, performance, workflow, and metagenomic-scale capabilities of eggNOG-mapper v2term·def
Research question — how to make orthology-based automated functional annotation more scalable, versatile, and accurate for large genomic and metagenomic data sets
Contribution — eggNOG-mapper v2term·def, with eggNOG v5term coverage, faster mapping, contig-level gene prediction, pairwise orthologyterm, domain prediction, and GFF decoration
Cross-disciplinary analysis
content-tags & facets:
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | large-scale transfer of gene functions and protein-domain annotations through taxonomically refined orthology — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | |
| claim-type | method — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | |
| method — evidence | — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” | |
| object of study — evidence | — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | |
| research question — evidence | — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | |
| contribution — evidence | — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | |
| mechanism | phylogeny-refined orthology transfer Precomputed OG phylogenies refine orthology assignments and limit transfer from putative in-paralogs. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | mechanisms[0] |
| mechanism | four-stage workflow Prodigal prediction is followed by sequence search, taxonomically scoped orthology inference, and annotation/domain transfer. — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | mechanisms[1] |
| mechanism | database expansion Synchronization with eggNOG v5 expands organism and orthologous-group coverage and phylogenetic resolution. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | mechanisms[2] |
| mechanism | memory-resident querying A memory-based mode reduces the impact of slow disk I/O on annotation rate. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | mechanisms[3] |
| mechanism | Prodigal ORF prediction Prodigal turns prokaryotic assembled contigs into protein sequences for annotation. — “Another major capability added to the new eggNOG-mapper workflow is predicting ORFs directly from assembled contigs ( fig. 1 A ). ORF detection, only available for prokaryotic assemblies, is performed using Prodigal ( Hyatt et al. 2010 ), which provides the protein sequences to be used by eggNOG-mapper for functional annotation. Prodigal modes (“normal,” “anonymous,” and “training”) as well as custom translation tables can be further chosen by the user” | mechanisms[4] |
| mechanism | iterative Diamond search Diamond's iterate mode applies successive sensitivity levels to improve search efficiency. — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | mechanisms[5] |
| mechanism | blastx-like mapping Diamond or MMseqs2 maps nucleotide sequencing reads without requiring prior assembly. — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | mechanisms[6] |
| mechanism | taxonomic-scope filtering Automatic or lineage-specific scopes limit the orthologs permitted to transfer annotations. — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | mechanisms[7] |
| mechanism | ortholog-domain transfer PFAM annotations are transferred from inferred orthologs at little computational cost. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | mechanisms[8] |
| mechanism | PFAM realignment refinement HMMER3 realignment refines transferred domains and applies GA thresholds and clan disambiguation. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | mechanisms[9] |
| extends | eggNOG-mapper version 1 — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | claim_relations[0] |
| builds-on | eggNOG v5 — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | claim_relations[1] |
| extends | orthology-based functional annotation — “Overall, eggNOG-mapper v2 provides a more efficient, versatile, and scalable automated functional annotation workflow than its predecessor. Standalone versions are available at GitHub ( https://github.com/eggnogdb/eggnog-mapper ), together with extensive documentation and usage examples ( https://github.com/eggnogdb/eggnog-mapper/wiki ). For convenience, an online service for the annotation of large genomic and metagenomic data sets is also available at http://eggnog-mapper.embl.de” | claim_relations[2] |
| sense | eggNOG-mapper v2 as an orthology-based automated functional-annotation workflow for genomic and metagenomic scale — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | senses[0] |
| sense | orthology-based transfer as using expected function retention in orthologs rather than similarity alone — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | senses[1] |
| sense | the eggNOG-mapper workflow as prediction, sequence search, orthology inference, and annotation transfer — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | senses[2] |
| sense | annotation rate as annotated queries per second — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | senses[3] |
| sense | Diamond mode as the default speed-memory compromise — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | senses[4] |
| sense | automatic taxonomic scope as the recommended mode for mixed metagenomes — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | senses[5] |
| sense | pairwise orthology reporting as a quick first pass rather than a substitute for precise inference — “Taking advantage of the rapid orthology assignments performed by eggNOG-mapper, it is now possible to report pairwise orthology relationships for each query against any of the genomes covered by eggNOG v5 ( fig. 1 C ). Although this feature is not intended to substitute more precise orthology prediction methods, it provides a very quick and simple “first-pass” approach to obtain pairwise relationships between query sequences and all eggNOG v5 organisms. Orthology reports can be further adjusted by specifying the target taxa and the type of orthologs to be reported (i.e., one-to-one, many-to-many)” | senses[6] |
| sense | direct PFAM transfer as low-cost domain prediction with residual false positives and negatives — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | senses[7] |
| sense | PFAM refinement as realignment of transferred domains to improve accuracy at low cost — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | senses[8] |
| sense | metagenomic-scale annotation as accepting contigs or reads and emitting functional profiles and decorated feature files — “On the other hand, this new version of eggNOG-mapper is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigal for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation tools, with only Prokka and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2 allows for the direct annotation of reads via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samples before assembling” | senses[9] |
| control | reannotation of 1.75 million randomly subsampled human-gut proteins — held — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” — rules out coverage gains caused only by a small hand-selected test set — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | controls[0] |
| control | Wilcoxon comparison of query alignment scores — held — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” — rules out treating apparent phylogenetic-resolution improvement as an untested visual difference — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | controls[1] |
| control | full de novo PFAM realignment used as the reference — held — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” — rules out evaluating transferred-domain accuracy without an independent computational reference — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” | controls[2] |
| control | random proteome, genome, and protein sets with stated CPU and RAM allocations — rules out unreported scale and compute conditions in runtime comparisons — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” | controls[3] |
| note | eggNOG-mapper version 1 — Version 2 adds databases, performance enhancements, gene prediction, pairwise orthology, domain discovery, and GFF decoration. — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | claim_relations[0] |
| weight | eggNOG-mapper version 1 — 0.95 — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | claim_relations[0] |
| note | eggNOG v5 — The workflow maps input proteins to and transfers annotations from eggNOG v5. — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | claim_relations[1] |
| note | orthology-based functional annotation — The paper contributes a more efficient, versatile, and scalable annotation workflow. — “Overall, eggNOG-mapper v2 provides a more efficient, versatile, and scalable automated functional annotation workflow than its predecessor. Standalone versions are available at GitHub ( https://github.com/eggnogdb/eggnog-mapper ), together with extensive documentation and usage examples ( https://github.com/eggnogdb/eggnog-mapper/wiki ). For convenience, an online service for the annotation of large genomic and metagenomic data sets is also available at http://eggnog-mapper.embl.de” | claim_relations[2] |
| weight | orthology-based functional annotation — 0.95 — “Overall, eggNOG-mapper v2 provides a more efficient, versatile, and scalable automated functional annotation workflow than its predecessor. Standalone versions are available at GitHub ( https://github.com/eggnogdb/eggnog-mapper ), together with extensive documentation and usage examples ( https://github.com/eggnogdb/eggnog-mapper/wiki ). For convenience, an online service for the annotation of large genomic and metagenomic data sets is also available at http://eggnog-mapper.embl.de” | claim_relations[2] |
| note | orthology —contrasted-with→ homology detection — Direct rather-than contrast; weight reflects the paper's generally-considered qualification. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concept_edges[0] |
| relation | orthology —contrasted-with→ homology detection — Function inference via orthology is contrasted with homology detection. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concept_edges[0] |
| weight | orthology —contrasted-with→ homology detection — 0.95 — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concept_edges[0] |
| note | orthologs —contrasted-with→ paralogs — Comparative expectation, not a necessary distinction. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concept_edges[1] |
| relation | orthologs —contrasted-with→ paralogs — Orthologs are expected to retain function more often than paralogs. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concept_edges[1] |
| weight | orthologs —contrasted-with→ paralogs — 0.9 — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concept_edges[1] |
| effect direction | underlying genome database —causes→ annotation coverage — increases — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[2] |
| note | underlying genome database —causes→ annotation coverage — Direct increase supported by a reannotation benchmark. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[2] |
| relation | underlying genome database —causes→ annotation coverage — The updated genome database increases annotation coverage. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[2] |
| weight | underlying genome database —causes→ annotation coverage — 0.95 — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[2] |
| effect direction | underlying genome database —causes→ phylogenetic resolution — increases — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[3] |
| note | underlying genome database —causes→ phylogenetic resolution — Direct increase supported by better alignment scores. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[3] |
| relation | underlying genome database —causes→ phylogenetic resolution — The updated genome database increases phylogenetic resolution. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[3] |
| weight | underlying genome database —causes→ phylogenetic resolution — 0.95 — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concept_edges[3] |
| note | eggNOG-mapper —contrasted-with→ Prokka — Direct measured performance comparison. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concept_edges[4] |
| relation | eggNOG-mapper —contrasted-with→ Prokka — eggNOG-mapper is benchmarked against Prokka and runs faster, especially on large metagenomic data sets. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concept_edges[4] |
| weight | eggNOG-mapper —contrasted-with→ Prokka — 0.95 — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concept_edges[4] |
| note | Diamond mode —contrasted-with→ MMseqs2 mode — Direct measured comparison; the numerical size regimes make full strength appropriate. — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | concept_edges[5] |
| relation | Diamond mode —contrasted-with→ MMseqs2 mode — Diamond and MMseqs2 modes are compared by query-set size, speed, sensitivity, and memory use. — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | concept_edges[5] |
| note | native results —distinct-from→ orthology-based predictions — Explicit independence statement. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concept_edges[6] |
| relation | native results —distinct-from→ orthology-based predictions — Full de novo computation obtains native results independent from orthology predictions. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concept_edges[6] |
| outcome | control — held — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | controls[0] |
| outcome | control — held — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | controls[1] |
| outcome | control — held — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” | controls[2] |
| relation | ORF detection → Prodigal — ORF detection is implemented using Prodigal. — “Another major capability added to the new eggNOG-mapper workflow is predicting ORFs directly from assembled contigs ( fig. 1 A ). ORF detection, only available for prokaryotic assemblies, is performed using Prodigal ( Hyatt et al. 2010 ), which provides the protein sequences to be used by eggNOG-mapper for functional annotation. Prodigal modes (“normal,” “anonymous,” and “training”) as well as custom translation tables can be further chosen by the user” | unhoused_relations[0] |
| relation | eggNOG-mapper → eggNOG database — eggNOG-mapper relies on the eggNOG database of orthologous groups. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | unhoused_relations[1] |
| weight | eggNOG-mapper → eggNOG database — 0.95 — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | unhoused_relations[1] |
| relation | precomputed phylogenies → putative in-paralogs — Precomputed phylogenies help minimize annotation transfer from putative in-paralogs. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | unhoused_relations[2] |
| weight | precomputed phylogenies → putative in-paralogs — 0.95 — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | unhoused_relations[2] |
| relation | database optimizations → annotation rate — Database optimizations, parallelization, and memory-based operation improve annotation rates. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | unhoused_relations[3] |
| weight | database optimizations → annotation rate — 0.95 — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | unhoused_relations[3] |
| relation | memory-based mode → slow I/O disk operations — The memory-based mode reduces the impact of slow I/O disk operations. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | unhoused_relations[4] |
| weight | memory-based mode → slow I/O disk operations — 0.95 — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | unhoused_relations[4] |
| relation | custom annotation databases → specific taxonomic groups — Custom annotation databases are constrained to specific taxonomic groups. — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | unhoused_relations[5] |
| weight | custom annotation databases → specific taxonomic groups — 0.95 — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | unhoused_relations[5] |
| relation | lineage-specific scopes → orthologs of unwanted lineages — Lineage-specific scopes prevent annotation transfer from orthologs of unwanted lineages. — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | unhoused_relations[6] |
| weight | lineage-specific scopes → orthologs of unwanted lineages — 0.95 — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | unhoused_relations[6] |
| relation | de novo PFAM domain annotation → false-positive and negative predictions — Realignment refinement reduces false-positive and false-negative domain predictions. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | unhoused_relations[7] |
| weight | de novo PFAM domain annotation → false-positive and negative predictions — 0.95 — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | unhoused_relations[7] |
| relation | annotation of reads → functional profiles of metagenomic samples — Direct read annotation can produce functional profiles before assembly. — “On the other hand, this new version of eggNOG-mapper is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigal for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation tools, with only Prokka and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2 allows for the direct annotation of reads via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samples before assembling” | unhoused_relations[8] |
| weight | annotation of reads → functional profiles of metagenomic samples — 0.7 — “On the other hand, this new version of eggNOG-mapper is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigal for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation tools, with only Prokka and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2 allows for the direct annotation of reads via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samples before assembling” | unhoused_relations[8] |
| no type for this | ORF detection → Prodigal — ORF detection is implemented using Prodigal. — “Another major capability added to the new eggNOG-mapper workflow is predicting ORFs directly from assembled contigs ( fig. 1 A ). ORF detection, only available for prokaryotic assemblies, is performed using Prodigal ( Hyatt et al. 2010 ), which provides the protein sequences to be used by eggNOG-mapper for functional annotation. Prodigal modes (“normal,” “anonymous,” and “training”) as well as custom translation tables can be further chosen by the user” | unhoused_relations[0] |
| no type for this | eggNOG-mapper → eggNOG database — eggNOG-mapper relies on the eggNOG database of orthologous groups. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | unhoused_relations[1] |
| no type for this | precomputed phylogenies → putative in-paralogs — Precomputed phylogenies help minimize annotation transfer from putative in-paralogs. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | unhoused_relations[2] |
| no type for this | database optimizations → annotation rate — Database optimizations, parallelization, and memory-based operation improve annotation rates. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | unhoused_relations[3] |
| no type for this | memory-based mode → slow I/O disk operations — The memory-based mode reduces the impact of slow I/O disk operations. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | unhoused_relations[4] |
| no type for this | custom annotation databases → specific taxonomic groups — Custom annotation databases are constrained to specific taxonomic groups. — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | unhoused_relations[5] |
| no type for this | lineage-specific scopes → orthologs of unwanted lineages — Lineage-specific scopes prevent annotation transfer from orthologs of unwanted lineages. — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | unhoused_relations[6] |
| no type for this | de novo PFAM domain annotation → false-positive and negative predictions — Realignment refinement reduces false-positive and false-negative domain predictions. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | unhoused_relations[7] |
| no type for this | annotation of reads → functional profiles of metagenomic samples — Direct read annotation can produce functional profiles before assembly. — “On the other hand, this new version of eggNOG-mapper is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigal for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation tools, with only Prokka and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2 allows for the direct annotation of reads via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samples before assembling” | unhoused_relations[8] |
| limitation | Automated gene-function annotation remains challenging at large genomic and metagenomic scales. — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | limitations[0] |
| limitation | Orthology delineation is computationally and algorithmically demanding, so many tools rely on homology instead. — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | limitations[1] |
| limitation | ORF detection is available only for prokaryotic assemblies. — “Another major capability added to the new eggNOG-mapper workflow is predicting ORFs directly from assembled contigs ( fig. 1 A ). ORF detection, only available for prokaryotic assemblies, is performed using Prodigal ( Hyatt et al. 2010 ), which provides the protein sequences to be used by eggNOG-mapper for functional annotation. Prodigal modes (“normal,” “anonymous,” and “training”) as well as custom translation tables can be further chosen by the user” | limitations[2] |
| limitation | Direct nucleotide translation assumes coding sequences begin in an open reading frame. — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | limitations[3] |
| limitation | HMMER3 is substantially slower than Diamond and MMseqs2 and requires large local databases. — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | limitations[4] |
| limitation | Pairwise orthology reports are a first-pass result and are not intended to replace more precise methods. — “Taking advantage of the rapid orthology assignments performed by eggNOG-mapper, it is now possible to report pairwise orthology relationships for each query against any of the genomes covered by eggNOG v5 ( fig. 1 C ). Although this feature is not intended to substitute more precise orthology prediction methods, it provides a very quick and simple “first-pass” approach to obtain pairwise relationships between query sequences and all eggNOG v5 organisms. Orthology reports can be further adjusted by specifying the target taxa and the type of orthologs to be reported (i.e., one-to-one, many-to-many)” | limitations[5] |
| limitation | Direct ortholog transfer of PFAM domains has residual false-positive and false-negative predictions. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | limitations[6] |
| limitation | Direct blastx-like annotation of reads is available but is not recommended in terms of accuracy. — “On the other hand, this new version of eggNOG-mapper is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigal for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation tools, with only Prokka and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2 allows for the direct annotation of reads via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samples before assembling” | limitations[7] |
Content tags
content-observations:
finding · quantity · variable · definition · key-term — grounded in the full text| finding | eggNOG-mapper v2 adds contig-level gene prediction, pairwise orthology reports, domain discovery, and GFF decoration. — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | content_tags.findings[0] |
| finding | The eggNOG v5 update spans 4.4 million orthologous groups and more than twice as many organisms as the previous version. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.findings[1] |
| finding | Reannotation of 1.75 million proteins increased annotation coverage by 3.23 percent, adding 56,569 annotations. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.findings[2] |
| finding | The expanded database produced significantly better query alignment scores than previous versions. — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.findings[3] |
| finding | Despite doubled database size, version 2 improves annotated queries per second by 16 percent on average. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | content_tags.findings[4] |
| finding | All computational enhancements together improve annotation rates by 608 percent on average over version 1. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | content_tags.findings[5] |
| finding | eggNOG-mapper runs faster than Prokka, especially on large metagenomic data sets. — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | content_tags.findings[6] |
| finding | Diamond iterate mode is twofold faster than MMseqs2 at one million queries and threefold faster at ten million. — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | content_tags.findings[7] |
| finding | MMseqs2 is faster than Diamond for query sets up to 100,000 at comparable sensitivity. — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | content_tags.findings[8] |
| finding | Lineage-specific scopes prevent functional transfer from unwanted lineages. — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | content_tags.findings[9] |
| finding | Direct PFAM transfer has an F1 score of 89.7 percent with little computational cost. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | content_tags.findings[10] |
| finding | Realignment refinement raises PFAM F1 to 98.9 percent while keeping computational cost low. — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | content_tags.findings[11] |
| finding | The workflow emits decorated GFF, tabular annotations, orthology assignments, and taxonomic placements. — “Regarding outputs, eggNOG-mapper is optimized for large-scale annotation jobs, producing GFF files decorated with functional annotations, tab-delimited files with the predicted functional terms per query, orthology assignment tables and taxonomic placements. Other tools provide similar output files (except for the orthology and taxonomic predictions), with only DRAM and MicrobiomeAnnotator providing further statistics and graphical outputs summarizing the results. Notably, Prokka can be set to produce Genbank and Sequin files compliant with the Genbank/ENA/DDJB format, thus facilitating the uploading of annotated genomes to public databases” | content_tags.findings[12] |
| quantity | 4.4 million OGs — orthologous groups in the synchronized eggNOG v5 genome database — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.quantities[0] |
| quantity | 1.75 million proteins — random human-gut catalog subsample used for reannotation — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.quantities[1] |
| quantity | 3.23% — increase in annotation coverage over version 1 — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.quantities[2] |
| quantity | 56,569 — newly annotated proteins in the coverage benchmark — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.quantities[3] |
| quantity | 16% — average annotation-rate gain despite doubled database size — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | content_tags.quantities[4] |
| quantity | 608% — average combined annotation-rate improvement over version 1 — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | content_tags.quantities[5] |
| quantity | 1,000,000 — query-set size where Diamond iterate is twice as fast as MMseqs2 — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | content_tags.quantities[6] |
| quantity | 10,000,000 — query-set size where Diamond iterate is three times faster than MMseqs2 — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | content_tags.quantities[7] |
| quantity | 100,000 — upper query-set size where MMseqs2 is faster at comparable sensitivity — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | content_tags.quantities[8] |
| quantity | 89.7% — F1 score for direct ortholog transfer of PFAM domains — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | content_tags.quantities[9] |
| quantity | 98.9% — F1 score after PFAM realignment refinement — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | content_tags.quantities[10] |
| variable | annotation coverage (output compared between database versions) — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.variables[0] |
| variable | alignment scores (phylogenetic-resolution comparison outcome) — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | content_tags.variables[1] |
| variable | annotation rate (performance outcome measured as annotated queries per second) — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | content_tags.variables[2] |
| variable | taxonomic scope (user-controlled constraint on annotation transfer) — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | content_tags.variables[4] |
| variable | target taxa (selection variable for pairwise orthology reports) — “Taking advantage of the rapid orthology assignments performed by eggNOG-mapper, it is now possible to report pairwise orthology relationships for each query against any of the genomes covered by eggNOG v5 ( fig. 1 C ). Although this feature is not intended to substitute more precise orthology prediction methods, it provides a very quick and simple “first-pass” approach to obtain pairwise relationships between query sequences and all eggNOG v5 organisms. Orthology reports can be further adjusted by specifying the target taxa and the type of orthologs to be reported (i.e., one-to-one, many-to-many)” | content_tags.variables[5] |
| variable | specificity (PFAM domain-prediction performance outcome) — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” | content_tags.variables[6] |
| variable | recall (PFAM domain-prediction performance outcome) — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” | content_tags.variables[7] |
| variable | F 1 score (combined PFAM domain-prediction performance outcome) — “Performance of eggNOG-mapper v2. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2 (blue) against eggNOG-mapper v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2 (blue) against Prokka (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2 (-s 2,4,6; black) against Diamond (iterate/sensitive mode; orange). ( D ) Specificity (Sp), recall (Re), and F 1 score, of PFAM domain annotation either from direct transference from orthologs, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotation, using either PFAM full de novo (brown) or realign to orthologs domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins” | content_tags.variables[8] |
| definition | eggNOG-mapper v2 — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | content_tags.definitions[0] |
| definition | annotation rate — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | content_tags.definitions[1] |
| definition | taxonomic scope — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | content_tags.definitions[2] |
| definition | lineage-specific scope — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | content_tags.definitions[3] |
| definition | orthology reports — “Taking advantage of the rapid orthology assignments performed by eggNOG-mapper, it is now possible to report pairwise orthology relationships for each query against any of the genomes covered by eggNOG v5 ( fig. 1 C ). Although this feature is not intended to substitute more precise orthology prediction methods, it provides a very quick and simple “first-pass” approach to obtain pairwise relationships between query sequences and all eggNOG v5 organisms. Orthology reports can be further adjusted by specifying the target taxa and the type of orthologs to be reported (i.e., one-to-one, many-to-many)” | content_tags.definitions[4] |
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 | automated functional annotation of genes — not defined in this paper's own words; computational assignment of gene functions — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[0] |
| concept | genomic and metagenomic workflows — not defined in this paper's own words; analysis workflows in which functional annotation is a fundamental step — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[1] |
| concept | eggNOG-mapper — not defined in this paper's own words; orthology-based functional-annotation software — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[2] |
| concept | tool — not defined in this paper's own words; software class assigned to eggNOG-mapper — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[3] |
| concept | precomputed orthology assignments — not defined in this paper's own words; stored orthology relationships used as the annotation basis — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[4] |
| concept | de novo gene prediction — not defined in this paper's own words; prediction of genes directly from raw contigs — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[5] |
| concept | pairwise orthology prediction — not defined in this paper's own words; prediction of query-to-genome ortholog relationships — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[6] |
| concept | protein domain discovery — not defined in this paper's own words; detection of protein domains — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[7] |
| concept | automated GFF decoration — not defined in this paper's own words; automatic addition of functional annotations to GFF records — “Even though automated functional annotation of genes represents a fundamental step in most genomic and metagenomic workflows, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapper, a tool for functional annotation based on precomputed orthology assignments, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2 now allows for: 1) de novo gene prediction from raw contigs, 2) built-in pairwise orthology prediction, 3) fast protein domain discovery, and 4) automated GFF decoration. eggNOG-mapper v2 is available as a standalone tool or as an online service at http://eggnog-mapper.embl.de” | concepts[8] |
| concept | orthology — not defined in this paper's own words; evolutionary relationship used to transfer functional information — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[9] |
| concept | homology detection — not defined in this paper's own words; similarity-based alternative to orthology inference — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[10] |
| concept | functional information — not defined in this paper's own words; gene-function annotations transferred between sequences — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[11] |
| concept | orthologs — not defined in this paper's own words; related sequences expected to retain function relatively often — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[12] |
| concept | paralogs — not defined in this paper's own words; related sequences used as the comparison class for function retention — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[13] |
| concept | eggNOG database — not defined in this paper's own words; database of orthologous groups used by eggNOG-mapper — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[14] |
| concept | orthologs groups — not defined in this paper's own words; eggNOG groups abbreviated OGs — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[15] |
| concept | precomputed phylogenies — not defined in this paper's own words; stored trees used to refine orthology assignments — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[16] |
| concept | putative in-paralogs — not defined in this paper's own words; candidate paralogs whose annotations the workflow seeks not to transfer — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[17] |
| concept | homology-based approaches — not defined in this paper's own words; annotation approaches used as the accuracy comparator — “Inference of gene function via orthology, rather than by homology detection, is generally considered the most reliable approach for transferring functional information between molecular sequences, as orthologs are expected to retain function more often than paralogs ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthology is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapper relies on the eggNOG database ( Huerta-Cepas et al. 2019 ) of orthologs groups (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogenies inferred for each OG to efficiently refine orthology assignments and therefore minimize the transferring of annotations from putative in-paralogs. The method was originally proven to provide more accurate predictions than homology-based approaches ( Huerta-Cepas et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2, a major upgrade featuring improvements in annotation coverage, overall performance, and program capabilities” | concepts[18] |
| concept | gene prediction stage — not defined in this paper's own words; workflow stage predicting proteins from contigs — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[19] |
| concept | Prodigal — not defined in this paper's own words; software used for prokaryotic protein prediction — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[20] |
| concept | protein prediction — not defined in this paper's own words; prediction of protein sequences from assembled contigs — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[21] |
| concept | assembled contigs — not defined in this paper's own words; assembled nucleotide inputs for ORF prediction — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[22] |
| concept | HMMER3 — not defined in this paper's own words; HMM sequence-search option — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[23] |
| concept | Diamond — not defined in this paper's own words; sequence-mapping option balancing speed and memory — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[24] |
| concept | MMseqs2 — not defined in this paper's own words; sequence-mapping option with a different size-speed regime — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[25] |
| concept | eggNOG v5 — not defined in this paper's own words; version-five genome and functional database — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[26] |
| concept | orthology inference stage — not defined in this paper's own words; workflow stage generating ortholog reports — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[27] |
| concept | protein annotations and domains — not defined in this paper's own words; functional and domain outputs transferred to queries — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[28] |
| concept | tabular and GFF files — not defined in this paper's own words; principal annotation output formats — “Workflow and new features of eggNOG-mapper v2. ( A ) The gene prediction stage uses Prodigal to perform protein prediction from assembled contigs. ( B ) During the search stage, HMMER3, Diamond, or MMseqs2 can be used to align the input proteins to eggNOG v5. ( C ) During the orthology inference stage, a report of orthologs is generated based on the desired taxonomic scope. ( D ) Finally, protein annotations and domains are transferred from orthologs to the queries and reported as tabular and GFF files” | concepts[29] |
| concept | underlying genome database — not defined in this paper's own words; updated eggNOG v5 genome collection — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concepts[30] |
| concept | annotation coverage — not defined in this paper's own words; proportion or count of proteins receiving annotations — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concepts[31] |
| concept | phylogenetic resolution — not defined in this paper's own words; resolution of sequence placement in the expanded database — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concepts[32] |
| concept | alignment scores — not defined in this paper's own words; query alignment-quality outcome — “The underlying genome database has been updated to be in sync with eggNOG v5, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverage and phylogenetic resolution, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverage (56,569 newly annotated proteins), compared with eggNOG-mapper version 1. The phylogenetic resolution was also improved, obtaining significantly better alignment scores for the query sequences than previous versions” | concepts[33] |
| concept | annotation rate — annotated queries per second — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concepts[34] |
| concept | database optimizations — not defined in this paper's own words; storage and query changes enabling faster and parallel operation — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concepts[35] |
| concept | memory-based mode — not defined in this paper's own words; mode that reduces dependence on slow disk input/output — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concepts[36] |
| concept | slow I/O disk operations — not defined in this paper's own words; disk bottleneck reduced by in-memory operation — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concepts[37] |
| concept | Prokka — not defined in this paper's own words; fast prokaryotic annotation comparator — “Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2 improves the annotation rate (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizations, allowing for faster queries and parallelization, and a new memory-based mode that significantly reduces the impact of slow I/O disk operations. Taken together, these changes improve annotation rates by 608% on average, with respect to eggNOG-mapper v1 ( fig. 2 A ). Compared with Prokka ( Seemann 2014 ), one of the fastest annotation tools available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapper runs faster, especially on large metagenomic data sets” | concepts[38] |
| concept | ORF detection — not defined in this paper's own words; prokaryotic open-reading-frame prediction from assembled contigs — “Another major capability added to the new eggNOG-mapper workflow is predicting ORFs directly from assembled contigs ( fig. 1 A ). ORF detection, only available for prokaryotic assemblies, is performed using Prodigal ( Hyatt et al. 2010 ), which provides the protein sequences to be used by eggNOG-mapper for functional annotation. Prodigal modes (“normal,” “anonymous,” and “training”) as well as custom translation tables can be further chosen by the user” | concepts[39] |
| concept | protein sequences — not defined in this paper's own words; Prodigal outputs passed to functional annotation — “Another major capability added to the new eggNOG-mapper workflow is predicting ORFs directly from assembled contigs ( fig. 1 A ). ORF detection, only available for prokaryotic assemblies, is performed using Prodigal ( Hyatt et al. 2010 ), which provides the protein sequences to be used by eggNOG-mapper for functional annotation. Prodigal modes (“normal,” “anonymous,” and “training”) as well as custom translation tables can be further chosen by the user” | concepts[40] |
| concept | custom translation tables — not defined in this paper's own words; user-selectable mappings from codons to amino acids — “Another major capability added to the new eggNOG-mapper workflow is predicting ORFs directly from assembled contigs ( fig. 1 A ). ORF detection, only available for prokaryotic assemblies, is performed using Prodigal ( Hyatt et al. 2010 ), which provides the protein sequences to be used by eggNOG-mapper for functional annotation. Prodigal modes (“normal,” “anonymous,” and “training”) as well as custom translation tables can be further chosen by the user” | concepts[41] |
| concept | sequence-mapping step — not defined in this paper's own words; initial mapping of input proteins to eggNOG v5 — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | concepts[42] |
| concept | Diamond mode — not defined in this paper's own words; default mapping mode balancing speed and memory — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | concepts[43] |
| concept | MMseqs2 mode — not defined in this paper's own words; mapping mode faster on smaller query sets at comparable sensitivity — “Additionally, we have broadened the options for the initial sequence-mapping step carried out by eggNOG-mapper ( fig. 1 B ). Now, Diamond, MMseqs2, and HMMER3 ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond mode provides the best balance between speed and memory consumption. EggNOG-mapper v2 also implements the new Diamond mode ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamond is twice as fast as MMseqs2 for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 mode provides faster results than Diamond ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity” | concepts[44] |
| concept | direct translation — not defined in this paper's own words; translation of coding nucleotide inputs before mapping — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | concepts[45] |
| concept | blastx-like mode — not defined in this paper's own words; six-frame-like mapping option for sequencing reads — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | concepts[46] |
| concept | sequencing reads — not defined in this paper's own words; nucleotide input for blastx-like search — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | concepts[47] |
| concept | frameshift option — not defined in this paper's own words; Diamond option for long error-prone reads — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | concepts[48] |
| concept | HMM-based searches — not defined in this paper's own words; slower searches aimed at distant relationships — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | concepts[49] |
| concept | distant homology relationships — not defined in this paper's own words; relationships potentially missed by Diamond and MMseqs — “When input data are nucleotide sequences, a direct translation is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamond and MMseqs2 can be executed in blastx-like mode, particularly useful when using sequencing reads as input data. For the annotation of long, error-prone sequencing reads, the frameshift option of Diamond can also be enabled. The HMMER3 mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searches might aid in the annotation of sequences for which distant homology relationships against the eggNOG v5 OGs cannot be detected by Diamond and MMseqs modes” | concepts[50] |
| concept | custom annotation databases — not defined in this paper's own words; databases constrained to chosen taxonomic groups — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | concepts[51] |
| concept | specific taxonomic groups — not defined in this paper's own words; domain or phylum scopes used to constrain databases — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | concepts[52] |
| concept | taxonomic scope — not defined in this paper's own words; lineage range permitted for functional transfer — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | concepts[53] |
| concept | automatic mode — not defined in this paper's own words; scope-selection mode recommended for mixed metagenomes — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | concepts[54] |
| concept | lineage-specific scopes — not defined in this paper's own words; scope controls excluding unwanted lineages — “Another new feature now available with eggNOG-mapper v2 is the possibility of creating custom annotation databases constrained to specific taxonomic groups. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scope ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic mode (recommended for mixed metagenomic data sets) to lineage-specific scopes (preventing transferring functional terms from orthologs of unwanted lineages)” | concepts[55] |
| concept | pairwise orthology relationships — not defined in this paper's own words; query-to-genome orthology report entries — “Taking advantage of the rapid orthology assignments performed by eggNOG-mapper, it is now possible to report pairwise orthology relationships for each query against any of the genomes covered by eggNOG v5 ( fig. 1 C ). Although this feature is not intended to substitute more precise orthology prediction methods, it provides a very quick and simple “first-pass” approach to obtain pairwise relationships between query sequences and all eggNOG v5 organisms. Orthology reports can be further adjusted by specifying the target taxa and the type of orthologs to be reported (i.e., one-to-one, many-to-many)” | concepts[56] |
| concept | first-pass — not defined in this paper's own words; quick preliminary pairwise-orthology approach — “Taking advantage of the rapid orthology assignments performed by eggNOG-mapper, it is now possible to report pairwise orthology relationships for each query against any of the genomes covered by eggNOG v5 ( fig. 1 C ). Although this feature is not intended to substitute more precise orthology prediction methods, it provides a very quick and simple “first-pass” approach to obtain pairwise relationships between query sequences and all eggNOG v5 organisms. Orthology reports can be further adjusted by specifying the target taxa and the type of orthologs to be reported (i.e., one-to-one, many-to-many)” | concepts[57] |
| concept | target taxa — not defined in this paper's own words; taxa selected for an orthology report — “Taking advantage of the rapid orthology assignments performed by eggNOG-mapper, it is now possible to report pairwise orthology relationships for each query against any of the genomes covered by eggNOG v5 ( fig. 1 C ). Although this feature is not intended to substitute more precise orthology prediction methods, it provides a very quick and simple “first-pass” approach to obtain pairwise relationships between query sequences and all eggNOG v5 organisms. Orthology reports can be further adjusted by specifying the target taxa and the type of orthologs to be reported (i.e., one-to-one, many-to-many)” | concepts[58] |
| concept | functional annotation sources — not defined in this paper's own words; controlled sources integrated into per-query reports — “In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2 offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sources, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologs ( Kanehisa et al. 2017 ); Gene Ontology labels ( Gene Ontology Consortium 2018 ); EC numbers, BiGG reactions ( Norsigian et al. 2020 ); CAZy terms ( Lombard et al. 2014 ); COG functional categories ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF file. Alternatively, eggNOG-mapper annotation reports can be used to decorate any custom GFF file” | concepts[59] |
| concept | Gene Ontology labels — not defined in this paper's own words; ontology terms included in functional reports — “In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2 offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sources, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologs ( Kanehisa et al. 2017 ); Gene Ontology labels ( Gene Ontology Consortium 2018 ); EC numbers, BiGG reactions ( Norsigian et al. 2020 ); CAZy terms ( Lombard et al. 2014 ); COG functional categories ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF file. Alternatively, eggNOG-mapper annotation reports can be used to decorate any custom GFF file” | concepts[60] |
| concept | EC numbers — not defined in this paper's own words; enzyme-commission identifiers included in reports — “In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2 offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sources, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologs ( Kanehisa et al. 2017 ); Gene Ontology labels ( Gene Ontology Consortium 2018 ); EC numbers, BiGG reactions ( Norsigian et al. 2020 ); CAZy terms ( Lombard et al. 2014 ); COG functional categories ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF file. Alternatively, eggNOG-mapper annotation reports can be used to decorate any custom GFF file” | concepts[61] |
| concept | BiGG reactions — not defined in this paper's own words; metabolic reaction identifiers included in reports — “In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2 offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sources, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologs ( Kanehisa et al. 2017 ); Gene Ontology labels ( Gene Ontology Consortium 2018 ); EC numbers, BiGG reactions ( Norsigian et al. 2020 ); CAZy terms ( Lombard et al. 2014 ); COG functional categories ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF file. Alternatively, eggNOG-mapper annotation reports can be used to decorate any custom GFF file” | concepts[62] |
| concept | CAZy terms — not defined in this paper's own words; carbohydrate-active enzyme annotations included in reports — “In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2 offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sources, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologs ( Kanehisa et al. 2017 ); Gene Ontology labels ( Gene Ontology Consortium 2018 ); EC numbers, BiGG reactions ( Norsigian et al. 2020 ); CAZy terms ( Lombard et al. 2014 ); COG functional categories ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF file. Alternatively, eggNOG-mapper annotation reports can be used to decorate any custom GFF file” | concepts[63] |
| concept | COG functional categories — not defined in this paper's own words; COG categories included in reports — “In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2 offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sources, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologs ( Kanehisa et al. 2017 ); Gene Ontology labels ( Gene Ontology Consortium 2018 ); EC numbers, BiGG reactions ( Norsigian et al. 2020 ); CAZy terms ( Lombard et al. 2014 ); COG functional categories ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF file. Alternatively, eggNOG-mapper annotation reports can be used to decorate any custom GFF file” | concepts[64] |
| concept | GFF file — not defined in this paper's own words; custom feature file that can be decorated with annotations — “In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2 offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sources, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologs ( Kanehisa et al. 2017 ); Gene Ontology labels ( Gene Ontology Consortium 2018 ); EC numbers, BiGG reactions ( Norsigian et al. 2020 ); CAZy terms ( Lombard et al. 2014 ); COG functional categories ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF file. Alternatively, eggNOG-mapper annotation reports can be used to decorate any custom GFF file” | concepts[65] |
| concept | PFAM — not defined in this paper's own words; protein-domain annotation source — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[66] |
| concept | SMART — not defined in this paper's own words; protein-domain annotation source — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[67] |
| concept | PFAM domain annotations — not defined in this paper's own words; PFAM predictions transferred or recomputed for queries — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[68] |
| concept | orthology-based predictions — not defined in this paper's own words; domain predictions transferred from inferred orthologs — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[69] |
| concept | false-positive and negative predictions — not defined in this paper's own words; errors possible under direct ortholog transfer — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[70] |
| concept | de novo PFAM domain annotation — not defined in this paper's own words; HMMER-based domain recomputation or refinement — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[71] |
| concept | refinement phase — not defined in this paper's own words; realignment step applied to transferred domain predictions — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[72] |
| concept | native results — not defined in this paper's own words; full-computation domain results independent of orthology — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[73] |
| concept | GA-based thresholds — not defined in this paper's own words; domain-specific significance thresholds — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[74] |
| concept | PFAM clan disambiguation — not defined in this paper's own words; resolution of overlapping domains from one clan — “Along with the functional terms annotated per query, this new version of eggNOG-mapper provides PFAM ( Mistry et al. 2021 ) and SMART ( Letunic et al. 2021 ) protein domain predictions. PFAM domain annotations are by default transferred from the inferred orthologs with very little impact on computational cost, but also with a small proportion of false-positive and negative predictions (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotation is also available at large scales, both as a refinement phase for the orthology-based predictions (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 score 98.9%, fig. 2 D ) or by full computation (obtaining native results independent from orthology predictions). When using the de novo approach, HMMER3 searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholds and PFAM clan disambiguation are automatically applied. Performance comparisons between the different modes are shown in figure 2 E” | concepts[75] |
| concept | broad annotation coverage — not defined in this paper's own words; cross-domain functional coverage of the eggNOG database — “EggNOG-mapper is tightly bound to the eggNOG database, which covers a wide range of prokaryotic and eukaryotic organisms, and provides normalized functional annotations from multiple sources (see previous sections). As eggNOG is based on a curated selection of representative species spanning the whole tree of life, it provides a broad annotation coverage while keeping redundancy low” | concepts[76] |
| concept | redundancy — not defined in this paper's own words; repeated reference content kept low by representative-species selection — “EggNOG-mapper is tightly bound to the eggNOG database, which covers a wide range of prokaryotic and eukaryotic organisms, and provides normalized functional annotations from multiple sources (see previous sections). As eggNOG is based on a curated selection of representative species spanning the whole tree of life, it provides a broad annotation coverage while keeping redundancy low” | concepts[77] |
| concept | annotation of reads — not defined in this paper's own words; direct blastx-like functional annotation before assembly — “On the other hand, this new version of eggNOG-mapper is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigal for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation tools, with only Prokka and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2 allows for the direct annotation of reads via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samples before assembling” | concepts[78] |
| concept | functional profiles of metagenomic samples — not defined in this paper's own words; preassembly metagenomic functional output — “On the other hand, this new version of eggNOG-mapper is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigal for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation tools, with only Prokka and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2 allows for the direct annotation of reads via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samples before assembling” | concepts[79] |
| concept | orthology assignment tables — not defined in this paper's own words; output tables unique to the eggNOG-mapper comparison — “Regarding outputs, eggNOG-mapper is optimized for large-scale annotation jobs, producing GFF files decorated with functional annotations, tab-delimited files with the predicted functional terms per query, orthology assignment tables and taxonomic placements. Other tools provide similar output files (except for the orthology and taxonomic predictions), with only DRAM and MicrobiomeAnnotator providing further statistics and graphical outputs summarizing the results. Notably, Prokka can be set to produce Genbank and Sequin files compliant with the Genbank/ENA/DDJB format, thus facilitating the uploading of annotated genomes to public databases” | concepts[80] |
| concept | taxonomic placements — not defined in this paper's own words; taxonomic output generated with annotations — “Regarding outputs, eggNOG-mapper is optimized for large-scale annotation jobs, producing GFF files decorated with functional annotations, tab-delimited files with the predicted functional terms per query, orthology assignment tables and taxonomic placements. Other tools provide similar output files (except for the orthology and taxonomic predictions), with only DRAM and MicrobiomeAnnotator providing further statistics and graphical outputs summarizing the results. Notably, Prokka can be set to produce Genbank and Sequin files compliant with the Genbank/ENA/DDJB format, thus facilitating the uploading of annotated genomes to public databases” | concepts[81] |
Full text — tagged
The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.
eggNOG-mapper v2term·def: Functional Annotation, Orthologyterm Assignments, and Domain Prediction at the Metagenomic Scale Abstract Even though automated functional annotation of genesterm represents a fundamental step in most genomic and metagenomic workflowsterm, it remains challenging at large scales. Here, we describe a major upgrade to eggNOG-mapperterm, a toolterm for functional annotation based on precomputed orthology assignmentsterm, now optimized for vast (meta)genomic data sets. Improvements in version 2 include a full update of both the genomes and functional databases to those from eggNOG v5term, as well as several efficiency enhancements and new features. Most notably, eggNOG-mapper v2term·def now allows for: 1) de novo gene predictionterm from raw contigs, 2) built-in pairwise orthology predictionterm, 3) fast protein domain discoveryterm, and 4) automated GFF decorationterm. eggNOG-mapper v2term·def is available as a standalone toolterm or as an online service at http://eggnog-mapperterm.embl.de .
Introduction Inference of gene function via orthologyterm, rather than by homology detectionterm, is generally considered the most reliable approach for transferring functional informationterm between molecular sequences, as orthologsterm are expected to retain function more often than paralogsterm ( Gabaldón and Koonin 2013 ; Glover et al. 2019 ). However, since delineating orthologyterm is highly demanding (both computationally and algorithmically), most automated methods rely on homology-based annotations ( Götz et al. 2008 ; Seemann 2014 ; Blum et al. 2021 ). EggNOG-mapperterm relies on the eggNOG databaseterm ( Huerta-CepaspersName et al. 2019 ) of orthologs groupsterm (OGs), covering thousands of bacterial, archaeal, and eukaryotic organisms. For this, it takes advantage of the precomputed phylogeniesterm inferred for each OG to efficiently refine orthologyterm assignments and therefore minimize the transferring of annotations from putative in-paralogsterm. The method was originally proven to provide more accurate predictions than homology-based approachesterm ( Huerta-CepaspersName et al. 2017 ), while preserving computational performance at the genomic and metagenomic scale. Here, we present eggNOG-mapper v2term·def, a major upgrade featuring improvements in annotation coverageterm, overall performance, and program capabilities ( fig. 1 ).
Workflow and new features of eggNOG-mapper v2term·def. ( A ) The gene prediction stageterm uses Prodigalterm to perform protein predictionterm from assembled contigsterm. ( B ) During the search stage, HMMER3term, Diamondterm, or MMseqs2term can be used to align the input proteins to eggNOG v5term. ( C ) During the orthology inference stageterm, a report of orthologsterm is generated based on the desired taxonomic scopeterm. ( D ) Finally, protein annotations and domainsterm are transferred from orthologsterm to the queries and reported as tabular and GFF filesterm.
Coverage and Performance Improvements The underlying genome databaseterm has been updated to be in sync with eggNOG v5term, spanning 4.4 million OGs and more than twice the number of organisms than in the previous version. This improvement increases annotation coverageterm and phylogenetic resolutionterm, particularly noticeable when analyzing large metagenomic data sets. For instance, the reannotation of 1.75 million proteins randomly subsampled from a human-gut metagenomic gene catalog ( Almeida et al. 2021 ) yielded a 3.23% increase in annotation coverageterm (56,569 newly annotated proteins), compared with eggNOG-mapperterm version 1. The phylogenetic resolutionterm was also improved, obtaining significantly better alignment scoresterm for the query sequences than previous versions (Wilcoxon test W = 1.2E + 12, P -value < 2.2E − 16). Moreover, although the underlying databases have doubled in size, eggNOG-mapper v2term·def improves the annotation rateterm (annotated queries per second) by 16% on average, compared with previous versions. The most important changes regarding computational enhancements relate to database optimizationsterm, allowing for faster queries and parallelization, and a new memory-based modeterm that significantly reduces the impact of slow I/O disk operationsterm. Taken together, these changes improve annotation ratesterm by 608% on average, with respect to eggNOG-mapperterm v1 ( fig. 2 A ). Compared with Prokkaterm ( Seemann 2014 ), one of the fastest annotation toolsterm available for prokaryotic genomes according to recent benchmarks ( Shaffer et al. 2020 ), eggNOG-mapperterm runs faster, especially on large metagenomic data sets ( fig. 2 B ).
Performance of eggNOG-mapper v2term·def. ( A ) average minutes to annotate input proteomes. EggNOG-mapper v2term·def (blue) against eggNOG-mapperterm v1 (red). ( B ) average minutes to annotate input genomes. EggNOG-mapper v2term·def (blue) against Prokkaterm (green). ( C ) average minutes (in log scale) to annotate input proteins. MMseqs2term (-s 2,4,6; black) against Diamondterm (iterate/sensitive mode; orange). ( D ) Specificityterm·var (Sp), recallterm·var (Re), and F 1 scoreterm·var, of PFAM domain annotationterm either from direct transference from orthologsterm, or after realignment. Full de novo realignment results were used as reference. ( E ) average minutes for PFAM domain annotationterm, using either PFAMterm full de novo (brown) or realign to orthologsterm domains (blue) modes. Benchmark setup: tests in ( A ) and ( B ) were done on 20 sets of 1–100 random proteomes ( A ) or genomes ( B ) from ( Almeida et al. 2021 ), and executed using 10 CPUs and 80 GB of RAM. Tests in ( C ) were done on 35 random sets of 10–10,000,000 proteins from Progenomes v2 ( Mende et al. 2020 ), using 30 CPUs and 240 GB of RAM. Tests in ( Dm ) and ( E ) as in ( C ), only for sets of 10–100,000 proteins.
ORF Prediction Another major capability added to the new eggNOG-mapperterm workflow is predicting ORFs directly from assembled contigsterm ( fig. 1 A ). ORF detectionterm, only available for prokaryotic assemblies, is performed using Prodigalterm ( Hyatt et al. 2010 ), which provides the protein sequencesterm to be used by eggNOG-mapperterm for functional annotation. Prodigalterm modes (“normal,” “anonymous,” and “training”) as well as custom translation tablesterm can be further chosen by the user.
Sequence Mapping Modes Additionally, we have broadened the options for the initial sequence-mapping stepterm carried out by eggNOG-mapperterm ( fig. 1 B ). Now, Diamondterm, MMseqs2term, and HMMER3term ( Mistry et al. 2013 ) modes are available, each recommended for different use cases. The default Diamond modeterm provides the best balance between speed and memory consumption. EggNOG-mapper v2term·def also implements the new Diamond modeterm ‘–iterate’, which iteratively applies different levels of sensitivity. Using the new –iterate mode, Diamondterm is twice as fast as MMseqs2term for query sets of 1,000,000, and three times faster for 10,000,000 sequences. The MMseqs2 modeterm provides faster results than Diamondterm ( fig. 2 C ) for query sets up to 100,000 sequences for comparable sensitivity. When input data are nucleotide sequences, a direct translationterm is done assuming they represent coding sequences starting in an open reading frame. Alternatively, both Diamondterm and MMseqs2term can be executed in blastx-like modeterm, particularly useful when using sequencing readsterm as input data. For the annotation of long, error-prone sequencing readsterm, the frameshift optionterm of Diamondterm can also be enabled. The HMMER3term mode is significantly slower than the other two and requires heavy databases to be downloaded. However, HMM-based searchesterm might aid in the annotation of sequences for which distant homology relationshipsterm against the eggNOG v5term OGs cannot be detected by Diamondterm and MMseqs modes.
Adjusting Taxonomic Scopesterm Another new feature now available with eggNOG-mapper v2term·def is the possibility of creating custom annotation databasesterm constrained to specific taxonomic groupsterm. For instance, users could easily create databases spanning only their domain or phylum of interest, therefore reducing computational times of subsequent annotation jobs. Moreover, the new version provides enhanced options to control the taxonomic scopeterm ( fig. 1 C ) used for transferring functional annotations, which can be adjusted from automatic modeterm (recommended for mixed metagenomic data sets) to lineage-specific scopesterm (preventing transferring functional terms from orthologsterm of unwanted lineages).
Orthology Reportsterm·def Taking advantage of the rapid orthologyterm assignments performed by eggNOG-mapperterm, it is now possible to report pairwise orthology relationshipsterm for each query against any of the genomes covered by eggNOG v5term ( fig. 1 C ). Although this feature is not intended to substitute more precise orthologyterm prediction methods, it provides a very quick and simple “first-passterm” approach to obtain pairwise relationships between query sequences and all eggNOG v5term organisms. Orthology reportsterm·def can be further adjusted by specifying the target taxaterm and the type of orthologsterm to be reported (i.e., one-to-one, many-to-many).
Annotation Sources In order to provide an integrated report of functional annotations per query, eggNOG-mapper v2term·def offers new annotation sources and improved reports ( fig. 1 D ). The functional annotation sourcesterm, which provide different levels of coverage ( supplementary fig. S1 , Supplementary Material online), are: predicted protein name; KEGG pathways, modules, and orthologsterm ( Kanehisa et al. 2017 ); Gene Ontology labelsterm ( Gene Ontology Consortium 2018 ); EC numbersterm, BiGG reactionsterm ( Norsigian et al. 2020 ); CAZy termsterm ( Lombard et al. 2014 ); COG functional categoriesterm ( Tatusov et al. 2000 ); eggNOG OGs; and free text descriptions at all taxonomic levels. Reports are generated in tab-delimited and/or XLSX file formats. Moreover, when ORF prediction mode is enabled, proteins used to annotate are reported in FASTA format, together with a functionally decorated GFF fileterm. Alternatively, eggNOG-mapperterm annotation reports can be used to decorate any custom GFF fileterm.
Protein Domain Annotations Along with the functional terms annotated per query, this new version of eggNOG-mapperterm provides PFAMterm ( Mistry et al. 2021 ) and SMARTterm ( LetunicpersName et al. 2021 ) protein domain predictions. PFAM domain annotationsterm are by default transferred from the inferred orthologsterm with very little impact on computational cost, but also with a small proportion of false-positive and negative predictionsterm (F1 score 89.7%, fig. 2 D ). Optionally, de novo PFAM domain annotationterm is also available at large scales, both as a refinement phaseterm for the orthology-based predictionsterm (thus keeping the computational cost very low, while eliminating the risk of false positives; F 1 scoreterm·var 98.9%, fig. 2 D ) or by full computation (obtaining native resultsterm independent from orthologyterm predictions). When using the de novo approach, HMMER3term searches are executed using in-memory mode for higher efficiency. Moreover, GA-based thresholdsterm and PFAM clan disambiguationterm are automatically applied. Performance comparisons between the different modes are shown in figure 2 E.
Features Comparison with Other Functional Annotation Toolsterm Although accuracy and performance are the main goal of eggNOG-mapper v2term·def, other practical differences might motivate its use, or not, compared with alternative software. Those practical issues include the type of input and output data, the functional sources used to annotate query sequences, and the target taxonomic scopeterm of each program.
EggNOG-mapperterm is tightly bound to the eggNOG databaseterm, which covers a wide range of prokaryotic and eukaryotic organisms, and provides normalized functional annotations from multiple sources (see previous sections). As eggNOG is based on a curated selection of representative species spanning the whole tree of life, it provides a broad annotation coverageterm while keeping redundancyterm low. Other toolsterm use smaller reference databases (e.g., SwissProt [ UniProt Consortium 2021 ]) for general functional annotations but incorporate niche-specific functional databases for specific purposes. For instance, Prokkaterm is suited for the annotation for prokaryotic-only genomes and includes the detection of tRNAs, transposases and signal peptides (which are not specifically targeted by eggNOG-mapperterm). DRAM ( Shaffer et al. 2020 ) and MicrobiomeAnnotator ( Ruiz-Perez et al. 2021 ) use broad annotation sources for microbial data sets such as KOFam/KEGG classification. Both provide optional mappings to large protein reference databases comparable to eggNOG (e.g., trEMBL or Uniref90), but at the cost of more computational resources and time ( Ruiz-Perez et al. 2021 ). In the case of DRAM, additional genomic features can be annotated, including rRNAs, tRNAs, peptidases, and carbohydrate-active enzymes. Mantis ( Queirós et al. 2021 ) and InterProScan allow users to annotate sequences based on multiple HMM-based database sources, including eggNOG, PFAMterm, or PANTHER ( Mi et al. 2010 ).
On the other hand, this new version of eggNOG-mapperterm is particularly focused on improving the annotation process at the metagenomic scale. For instance, it incorporates the possibility of using Prodigalterm for the automatic prediction of open reading frames (ORFs) out of raw contigs. Those options are not commonly available in other annotation toolsterm, with only Prokkaterm and DRAM providing similar capabilities. Moreover, eggNOG-mapper v2term·def allows for the direct annotation of readsterm via blastx-like searches, which is not a recommended mode in terms of accuracy, but it offers the possibility of obtaining functional profiles of metagenomic samplesterm before assembling.
Regarding outputs, eggNOG-mapperterm is optimized for large-scale annotation jobs, producing GFF filesterm decorated with functional annotations, tab-delimited files with the predicted functional terms per query, orthology assignment tablesterm and taxonomic placementsterm. Other toolsterm provide similar output files (except for the orthologyterm and taxonomic predictions), with only DRAM and MicrobiomeAnnotator providing further statistics and graphical outputs summarizing the results. Notably, Prokkaterm can be set to produce Genbank and Sequin files compliant with the Genbank/ENA/DDJB format, thus facilitating the uploading of annotated genomes to public databases.
Conclusions Overall, eggNOG-mapper v2term·def provides a more efficient, versatile, and scalable automated functional annotation workflow than its predecessor. Standalone versions are available at GitHub ( https://github.com/eggnogdb/eggnog-mapperterm ), together with extensive documentation and usage examples ( https://github.com/eggnogdb/eggnog-mapperterm/wiki ). For convenience, an online service for the annotation of large genomic and metagenomic data sets is also available at http://eggnog-mapperterm.embl.de .
Supplementary Material Supplementary data are available at Molecular Biology and Evolution online.
Supplementary Material Click here for additional data file.