AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models Mihaly Varadi Stephen Anyango Mandar Deshpande Sreenath Nair Cindy Natassia Galabina Yordanova David Yuan Oana Stroe Gemma Wood Agata Laydon Augustin Žídek Tim Green Kathryn Tunyasuvunakool Stig Petersen John Jumper Ellen Clancy Richard Green Ankur Vora Mira Lutfi Michael Figurnov Andrew Cowie Nicole Hobbs Pushmeet Kohli Gerard Kleywegt Ewan Birney Demis Hassabis Sameer Velankar

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

 AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models 2021 Nucleic Acids Research 10.1093/nar/gkab1061 varadi2021-alphafold-protein-structure-database varadi2021-alphafold-protein-structure-database.txt, the project's staged copy 9ca231f5c0c457782577532bc54185fa4e457ce2 cc-by structural-biology

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

 118 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them the AlphaFold Protein Structure Database, its predicted structures, confidence outputs, and access mechanisms how to expand protein-sequence-space structural coverage and deliver high-accuracy predicted structures with interpretable confidence information an openly accessible database initially providing more than 360,000 AlphaFold structures, confidence estimates, and bulk, API, and interactive access method the widening gap between known protein sequences and experimentally determined protein structures system-or-tool-buildingsystem-or-tool-building Artificial-intelligence techniques predict a protein's structure computationally from its amino-acid sequence.sequence-to-structure prediction Predicted per-residue scores on the lDDT-C-alpha metric provide a 0-to-100 local confidence estimate.pLDDT local-confidence scoring lDDT awards high scores to locally accurate regions even when a whole prediction cannot be globally aligned well.local-distance-difference assessment Residue-conditioned positional error estimates indicate whether domains have reliable relative positions and orientations.PAE pairwise-confidence assessment Apache Solr loads and indexes archived metadata so users can search on AlphaFold DB web pages.metadata indexing and search Protein pages combine 3D coordinates with pLDDT colour bands and a PAE plot to expose local and pairwise uncertainty.confidence-aware structure visualization Experimentally determining high-resolution structures remains labour-intensive, trial-and-error dependent, and unable to keep pace with known protein sequences. The initial prediction set excludes sequences with non-standard amino acids and does not provide multiple isoforms. Model confidence can vary substantially along a chain, so structural features should not be interpreted before examining confidence measures. High PAE values mean that relative positions and orientations of domains are unreliable and should not be assigned biological or structural relevance. The first release covers only 21 model-organism proteomes; broader UniRef90 coverage is presented as future work. AlphaFold Protein Structure DatabaseAlphaFold Protein Structure Database AlphaFold DBAlphaFold DB openly accessible, extensive databaseopenly accessible, extensive database high-accuracy protein-structure predictionshigh-accuracy protein-structure predictions structural coveragestructural coverage protein-sequence spaceprotein-sequence space predicted atomic coordinatespredicted atomic coordinates model-confidence estimatesmodel-confidence estimates predicted aligned errorspredicted aligned errors Universal Protein ResourceUniversal Protein Resource UniProtUniProt Protein Data BankProtein Data Bank PDBPDB Artificial IntelligenceArtificial Intelligence AlphaFoldAlphaFold AI systemAI system CASPCASP Critical Assessment of Structure PredictionsCritical Assessment of Structure Predictions biennial challengebiennial challenge UniRef90UniRef90 UniProt reference proteomeUniProt reference proteome per-residue confidence estimatesper-residue confidence estimates confidence measureconfidence measure pLDDTpLDDT predicted per-residue scorespredicted per-residue scores lDDTlDDT local accuracy of a predictionlocal accuracy of a prediction model confidencemodel confidence Predicted Aligned ErrorPredicted Aligned Error PAEPAE expected positional errorexpected positional error very low confidence pLDDT scoresvery low confidence pLDDT scores high propensities for intrinsic disorderhigh propensities for intrinsic disorder Apache SolrApache Solr application programming interfaceapplication programming interface APIAPI UniProt accessionUniProt accession AlphaFold DB web pagesAlphaFold DB web pages 3D molecular viewer3D molecular viewer pairwise confidence predictionpairwise confidence prediction reliability of relative domain positions and orientationsreliability of relative domain positions and orientations structural bioinformatics researchstructural bioinformatics research experimentally determined structuresexperimentally determined structures training of AlphaFoldtraining of AlphaFold very low confidence pLDDT scores → high propensities for intrinsic disorderVery low confidence pLDDT scores correlate with high propensities for intrinsic disorder. Apache Solr → AlphaFold DB web pagesApache Solr indexes metadata to enable searching on the AlphaFold DB web pages. pairwise confidence prediction → reliability of relative domain positions and orientationsThe pairwise confidence prediction helps assess the reliability of relative domain positions and orientations. experimentally determined structures → training of AlphaFoldExperimentally determined structures archived in PDB enabled the training of AlphaFold. AlphaFold DB as an openly accessible database of predicted protein structures and their confidence outputs structural coverage as coverage of known protein-sequence space by experimental or predicted structures AlphaFold as a sequence-to-structure AI system evaluated through the CASP benchmark pLDDT as a per-residue estimate of local model confidence derived from the lDDT-C-alpha metric lDDT as a local-accuracy metric that can score well-predicted regions despite poor global alignment PAE as residue-conditioned expected positional error used to evaluate relative domain placement AlphaFold DB access as bulk FTP, programmatic API, and protein-specific interactive pages pLDDT visualization as confidence-band colouring of residues in a 3D molecular viewer pairwise confidence as a PAE-coloured assessment of relative positions, orientations, and global topology experimental and predicted structure resources as complementary responses to a large sequence-structure coverage gap builds-on → AlphaFold v2.0 of DeepMindnote: The paper states that the database is powered by AlphaFold v2.0. extends → the structural coverage available from experimentally determined protein structuresnote: The database is presented as an unprecedented expansion of known protein-sequence-space coverage. AlphaFold Protein Structure Database —same-as→ AlphaFold DBthe relation itself is separately evidenced · strength of the relation: 1 · note: The parenthetical naming equivalence is categorical. · the relation stated: AlphaFold DB is the short name given in apposition to AlphaFold Protein Structure Database.AlphaFold Protein Structure DatabaseAlphaFold DB AlphaFold Protein Structure Database —is-a→ openly accessible, extensive databasethe relation itself is separately evidenced · strength of the relation: 1 · note: The paper directly assigns the resource to this class. · the relation stated: The AlphaFold Protein Structure Database is an openly accessible, extensive database.AlphaFold Protein Structure Databaseopenly accessible, extensive database Universal Protein Resource —same-as→ UniProtthe relation itself is separately evidenced · strength of the relation: 1 · note: The naming equivalence is explicit and categorical. · the relation stated: UniProt is the parenthetical short name of the Universal Protein Resource.Universal Protein ResourceUniProt Protein Data Bank —same-as→ PDBthe relation itself is separately evidenced · strength of the relation: 1 · note: The naming equivalence is explicit and categorical. · the relation stated: PDB is the parenthetical short name of the Protein Data Bank.Protein Data BankPDB Universal Protein Resource —contrasted-with→ Protein Data Bankthe relation itself is separately evidenced · strength of the relation: 0.95 · note: The 'while' clause makes an unhedged resource-and-coverage contrast. · the relation stated: UniProt's sequence holdings are contrasted with PDB's much smaller experimental-structure holdings.Universal Protein ResourceProtein Data Bank AlphaFold —is-a→ AI systemthe relation itself is separately evidenced · strength of the relation: 1 · note: This is a direct categorical class statement. · the relation stated: AlphaFold is an AI system developed by DeepMind.AlphaFoldAI system CASP —same-as→ Critical Assessment of Structure Predictionsthe relation itself is separately evidenced · strength of the relation: 1 · note: The paper gives the names in apposition. · the relation stated: Critical Assessment of Structure Predictions is the expanded name of CASP.CASPCritical Assessment of Structure Predictions CASP —is-a→ biennial challengethe relation itself is separately evidenced · strength of the relation: 1 · note: This is a direct categorical class statement. · the relation stated: CASP is a biennial challenge for research groups.CASPbiennial challenge confidence measure —same-as→ pLDDTthe relation itself is separately evidenced · strength of the relation: 1 · note: The paper explicitly names the measure. · the relation stated: The per-residue confidence measure is called pLDDT.confidence measurepLDDT pLDDT —operationalised-as→ predicted per-residue scoresthe relation itself is separately evidenced · strength of the relation: 0.95 · note: The score correspondence is directly stated; 0.95 preserves a small distinction between the named metric and its score basis. · the relation stated: pLDDT corresponds to predicted per-residue scores on the lDDT-C-alpha metric.pLDDTpredicted per-residue scores Predicted Aligned Error —same-as→ PAEthe relation itself is separately evidenced · strength of the relation: 1 · note: The naming equivalence is explicit and categorical. · the relation stated: PAE is the parenthetical short name of Predicted Aligned Error.Predicted Aligned ErrorPAE PAE —operationalised-as→ expected positional errorthe relation itself is separately evidenced · strength of the relation: 0.95 · note: The paper directly specifies the measured quantity; the relation is gradable rather than taxonomic. · the relation stated: PAE indicates expected positional error under alignment on a residue and is measured in angstroms.PAEexpected positional error application programming interface —same-as→ APIthe relation itself is separately evidenced · strength of the relation: 1 · note: The naming equivalence is explicit and categorical. · the relation stated: API is the parenthetical short name of application programming interface.application programming interfaceAPI The experimental-structure archive covers only a small fraction of the known protein-sequence space. The first AlphaFold DB release contains more than 360,000 predicted structures across 21 model-organism proteomes. AlphaFold was recognized by the CASP14 organizers as a solution to the protein-structure-prediction problem. pLDDT reflects local confidence, including in cases where individual domains are reliable but their relative positions are not. Very low pLDDT confidence scores correlate with intrinsic disorder propensity. Low cross-domain PAE indicates well-defined relative placement, while high PAE indicates unreliable placement. The database exposes predictions through bulk FTP, a public API, and protein-specific interactive pages. The pairwise PAE plot is linked to the 3D viewer to locate domains whose relative geometry is predicted confidently. The authors project expansion from the initial release to more than 100 million representative UniRef90 structures. an openly accessible database of high-accuracy protein-structure predictions and associated confidence outputsAlphaFold Protein Structure Database a DeepMind AI system that predicts protein structures from amino-acid sequencesAlphaFold a biennial challenge testing prediction accuracy against experimental structure dataCASP the model's predicted per-residue score on the lDDT-C-alpha metricpLDDT a protein-structure-prediction metric designed to assess local prediction accuracylDDT expected positional error at one residue when predicted and actual structures are aligned on anotherPredicted Aligned Error almost 220 million — unique protein sequences archived by UniProt just over 180 000 — 3D structures held by PDB at publication over 55 000 — distinct proteins represented by those PDB structures over 360 000 — predicted structures in the initial AlphaFold DB release 21 — model-organism proteomes in the first database release 0 to 100 — scale for per-residue confidence estimates 31.75 — angstrom cap applied to PAE values >100 million — planned predicted structures for representative UniRef90 sequences amino-acid sequenceamino-acid sequence pLDDTpLDDT PAE valuesPAE values UniProt accessionUniProt accession model confidencemodel confidence

AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models

Abstract

The AlphaFold Protein Structure Database (AlphaFold DB, https://alphafold.ebi.ac.uk ) is an openly accessible, extensive database of high-accuracy protein-structure predictions. Powered by AlphaFold v2.0 of DeepMind, it has enabled an unprecedented expansion of the structural coverage of the known protein-sequence space. AlphaFold DB provides programmatic access to and interactive visualization of predicted atomic coordinates, per-residue and pairwise model-confidence estimates and predicted aligned errors. The initial release of AlphaFold DB contains over 360,000 predicted structures across 21 model-organism proteomes, which will soon be expanded to cover most of the (over 100 million) representative sequences from the UniRef90 data set.

INTRODUCTION

Proteins are essential macromolecules with vital biological functions and, thus, are involved in a wide range of research activities and medical and biotechnological applications, from fighting infectious diseases to tackling environmental pollution ( 1 , 2 ). Knowledge of the three-dimensional (3D) arrangement of the atoms of a protein can provide essential clues to understanding the roles and mechanisms underpinning protein functions ( 3 , 4 ). However, while the Universal Protein Resource (UniProt) archives almost 220 million unique protein sequences, the Protein Data Bank (PDB) holds only just over 180 000 3D structures for over 55 000 distinct proteins, thus severely limiting the coverage of the sequence space to support biomolecular research globally ( 5–7 ).

Achieving a higher coverage of the sequence space with experimentally determined high-resolution structures is very labour-intensive. It often requires a lot of trial and error, for example, to find suitable constructs or conditions under which a protein is amenable to crystallization. Although recent advances in the field of electron cryo-microscopy and hybrid and integrative methods (I/HM) for structure determination have accelerated the pace of structure determination, the gap between known protein sequences and experimental protein structures continues to expand ( 6 , 8 ).

One way to close this gap is to predict the structures of millions of proteins. Increasingly, researchers deploy Artificial Intelligence (AI) techniques to predict a protein's structure computationally from its amino-acid sequence alone ( 9–11 ).

AlphaFold is an AI system developed by DeepMind that makes state-of-the-art predictions of protein structures from their amino-acid sequences ( 9 ). CASP (Critical Assessment of Structure Predictions) is a biennial challenge for research groups to test the accuracy of their predictions against actual experimental data. In 2020, the organizers of the CASP14 benchmark recognized AlphaFold as a solution to the protein–structure–prediction problem ( 12 ). The unprecedented accuracy and speed of AlphaFold allowed the creation of an extensive database of structure predictions at a large scale. It will enable biologists to obtain structural models for almost any protein sequence, changing how they tackle research questions and accelerate their projects. The methodology of AlphaFold and insights gained from the predictions for the complete human proteome have been described recently ( 9 , 13 ).

We present the AlphaFold Protein Structure Database (AlphaFold DB, https://alphafold.ebi.ac.uk ), a new data resource created in partnership between DeepMind and the EMBL-European Bioinformatics Institute (EMBL-EBI). We have created AlphaFold DB to make structure predictions freely available to the scientific community at a large scale. The first release described here covers the human proteome and those of 20 other model organisms (Table 1 ). In the coming months, we plan to have expanded the database to cover a large proportion of all catalogued proteins (over 130 million cluster representatives from UniRef90).

Structural predictions for complete proteomes in AlphaFold DB

IMPLEMENTATION

The initial version of AlphaFold DB contains over 360 000 predicted structures, corresponding meta-information and confidence metrics. All the data are publicly accessible through a cloud-based infrastructure. We have attempted to predict most sequences in the UniProt reference proteome in the 16–2700 amino acid length range (as well as 1400-residue fragments to cover longer human proteins) for the organisms currently covered. We excluded sequences that contain non-standard amino acids. We do not provide multiple isoforms at this point.

The predicted structures contain atomic coordinates and per-residue confidence estimates on a scale from 0 to 100, with higher scores corresponding to higher confidence. This confidence measure is called pLDDT and corresponds to the model's predicted per-residue scores on the lDDT-Cα metric ( 14 ). lDDT is a pre-existing metric used in the protein structure prediction field. A key motivation behind lDDT is to assess the local accuracy of a prediction, awarding a high score for regions that are well-predicted even if the entire prediction cannot be aligned well to the true structure. This is particularly important for evaluating multi-domain predictions where the individual domains may be largely accurate while their relative position is not. As a confidence metric based on lDDT, pLDDT also reflects local confidence in the structure, and should be used, for example, to assess confidence within a single domain. Several other protein structure prediction resources also use lDDT-based metrics ( 15 , 16 ). AlphaFold DB stores these values in the B-factor fields of the mmCIF and PDB files available for download and uses confidence bands based on these values to colour-code the residues of the models in the 3D structure viewer on the structure pages. Residues with pLDDT ≥ 90 have very high model confidence, while residues with 90 > pLDDT ≥ 70 are classified as confident. Residues with 70 > pLDDT ≥ 50 have low confidence, and residues with pLDDT < 50 correspond to very low confidence ( 13 ). It was recently described that very low confidence pLDDT scores correlate with high propensities for intrinsic disorder ( 17 ).

The Predicted Aligned Error (PAE) is another output of the AlphaFold system. It indicates the expected positional error at residue x if the predicted and actual structures are aligned on residue y (using the Cα, N and C atoms). PAEs are measured in Ångströms and capped at 31.75 Å. Scientists can use these values to assess the confidence in the relative position and orientation of different parts of the model (e.g. two domains). For residues x and y in two different domains, if the PAE values ( x , y ) are low, AlphaFold predicts the domains to have well-defined relative positions and orientations. If the PAE values are high, then the relative position and orientation of the two domains are unreliable, and users should not attach biological or structural relevance to these. Note that the PAE is asymmetric; therefore, there can be a difference between the PAE values for ( x, y ) and ( y, x ), for example, between loop regions with highly uncertain orientation.

Data archival

AlphaFold DB archives and provides access to the atomic coordinates in PDB and mmCIF formats, PAEs in JSON format and corresponding metadata in JSON format. While the coordinates and the PAE files are directly accessible through URLs, we load and index the metadata using the open-source search platform Apache Solr ( https://solr.apache.org/ ) to enable users to search on the AlphaFold DB web pages. The data files in the archive are versioned, and previous snapshots of the data will be available via FTP, but the web pages will always display the latest version.

Data access

AlphaFold DB provides predictions through several data-access mechanisms: (i) bulk downloads via FTP; (ii) programmatic access via an application programming interface (API); (iii) download and interactive visualization of individual predictions on protein-specific web pages keyed on UniProt accessions.

For bulk downloading data from AlphaFold DB, users can access the uncompressed archive files (.tar) of compressed PDB/mmCIF files (.gz) per reference proteome from the EMBL-EBI public FTP area at ftp://ftp.ebi.ac.uk/pub/databases/alphafold . This area contains the TAR files and a JSON file that provides meta-information, describing the species names (scientific and common), the reference proteome identifiers, the number of predicted structures, and the archives’ sizes. The same information and files are also available from the Bulk Download page of AlphaFold DB at https://alphafold.ebi.ac.uk/download .

We provide access to all entries through a public API endpoint, keyed on a UniProt accession. For example, the endpoint https://alphafold.ebi.ac.uk/api/prediction/Q92793 allows access to all the meta-information and the URLs of all the archived data files related to the human CREB-binding protein. UniProt ( 5 ), Pfam ( 18 ), InterPro ( 19 ) and PDBe-KB ( 7 ) use this API to display AlphaFold models on their web pages.

AlphaFold DB provides graphical access to and interactive visualization of all the predictions and meta-information for the broader scientific community through web pages. These pages contain all the available information for a protein of interest, keyed by its UniProt accession. They allow users to analyse the prediction and download the corresponding model files (in PDB and mmCIF formats) and PAE files (in JSON format).

AlphaFold DB web pages

AlphaFold DB provides convenient access to its predictions through a set of web pages ( https://alphafold.ebi.ac.uk ). These pages contain an introduction to the AlphaFold system, address the most frequent questions, enable bulk download of complete proteomes, and offer a search engine for finding pages specific to a protein of interest (Figure 1 ). Users can search by gene name, protein name, UniProt accession or organism name. The search results can be filtered, for example, only to show human proteins.

Searching AlphaFold DB. AlphaFold DB provides a search engine to find proteins of interest based on gene or protein name, UniProt accession or organism name. The search results can be filtered if required and clicking on a protein name leads to the relevant protein-specific entry page.

Each protein has a dedicated structure page that shows basic information (drawn from UniProt ( 5 ) and PDBe ( 6 )) and three separate outputs of the AlphaFold model. The first two outputs are the 3D coordinates and the per-residue confidence metric pLDDT, which is used to colour the residues of the model in the integrated 3D molecular viewer, Mol* ( 20 ). Model confidence can vary significantly along a chain, making it essential to analyse the confidence measures before interpreting structural features. The lower confidence bands appear to correlate well with backbone flexibility and intrinsic disorder ( 13 ) (Figure 2 ).

Meta-information and 3D visualization of the AlphaFold structure predictions. The protein-specific web pages display essential metadata for the protein of interest, such as known biological functions and cross-references to UniProt and PDBe-KB. Users can download the predicted models in PDB and mmCIF format, and an interactive molecular viewer visualizes the structure, coloured by the per-residue pLDDT confidence measure.

The third output is a pairwise confidence prediction, which helps to assess the reliability of relative domain positions and orientations as well as the global topology of the protein (Figure 3 ). The plot is coloured by the pairwise PAE values and it helps users to identify which domains have reliably predicted positions and orientations relative to one another, where dark green indicates high confidence. Selecting a region in the plot also highlights the corresponding part of the sequence in the 3D viewer.

Visualization of Predicted Aligned Errors. Protein-specific pages contain an interactive 2D plot of the PAE values. This tool interacts with the 3D molecular viewer to facilitate the identification of domains whose relative positions and orientations AlphaFold predicts with confidence. In this example ( https://alphafold.ebi.ac.uk/entry/Q93074 ), AlphaFold has high confidence in the relative position of domains at residues 1–500 (green) and residues 1200–1700 (blue), but not with the region between 500–1200 (orange) nor the C-terminus.

CONCLUSION AND OUTLOOK

Since the mid-1950s, the scientific community has been using ever-more advanced experimental methods to determine over 180 000 structures of proteins, nucleic acids, and complexes in atomic detail, and archive them in the PDB, the single worldwide archive of experimental macromolecular structure data managed by the wwPDB consortium ( 21 ). This collective body of work has vastly improved our understanding of many fundamental processes in health and disease, as evidenced in part by many Nobel Prizes for structures deposited in the PDB. Recently, determining the structure of the SARS-CoV-2 viral proteins enabled scientists to understand how it operates and to identify potential treatments and develop new vaccines ( 3 ). However, figuring out the exact structure of a protein remains an expensive and often time-consuming process. Thus, we only know the 3D structure of a tiny fraction of all proteins currently known to science.

The first release of AlphaFold DB contains over 360 000 predicted structures from 21 model-organism proteomes. Having access to these highly accurate models will greatly impact biology, from enabling structure-based drug design to providing data for high-throughput structural bioinformatics research that will tackle fundamental biological questions. We have already gained some invaluable insights from the predictions of the human proteome ( 13 ).

In the coming months, we will expand AlphaFold DB to provide structural predictions to include additional proteomes to support research in neglected diseases and to cover the set of highly annotated proteins in SwissProt, taking the number of structures available to >1 million. This will be followed by another update in 2022 to include structures for most representative sequences from the UniRef90 data set (>100 million structures). Future updates will also aim to overlay annotations onto the predicted structures and display this information on 2D sequence-feature viewers. AlphaFold DB will enable biomedical scientists to use 3D models of protein structures as a core tool, driving research and innovation across multiple fields by providing open access to an ever-growing number of predicted structures.

DATA AVAILABILITY

All the AlphaFold predictions are publicly available through multiple data-access mechanisms. Coordinate files in PDB and mmCIF formats are available in TAR archives per proteome through FTP at ftp://ftp.ebi.ac.uk/pub/databases/alphafold . Meta-information and URLs to individual UniProt accessions are available via a public API endpoint. For example, https://alphafold.ebi.ac.uk/api/prediction/Q92793 provides all the information for UniProt accession Q92793 ( https://www.alphafold.ebi.ac.uk/entry/Q92793 ).

ACKNOWLEDGEMENTS

We would like to acknowledge all the scientists who contributed valuable feedback throughout the development of this data resource. We would also like to recognize the contributions of all the structural biologists whose experimentally determined structures, archived in the PDB, enabled the training of AlphaFold. We further acknowledge the work of the public protein sequence archives such as the UniProt consortium, BFD and MGnify in collecting and organizing protein-sequence data which was used for predicting structures.

Contributor Information

Mihaly Varadi, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Stephen Anyango, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Mandar Deshpande, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Sreenath Nair, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Cindy Natassia, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Galabina Yordanova, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

David Yuan, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Oana Stroe, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Gemma Wood, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Agata Laydon, DeepMind, London, UK.

Augustin Žídek, DeepMind, London, UK.

Tim Green, DeepMind, London, UK.

Kathryn Tunyasuvunakool, DeepMind, London, UK.

Stig Petersen, DeepMind, London, UK.

John Jumper, DeepMind, London, UK.

Ellen Clancy, DeepMind, London, UK.

Richard Green, DeepMind, London, UK.

Ankur Vora, DeepMind, London, UK.

Mira Lutfi, DeepMind, London, UK.

Michael Figurnov, DeepMind, London, UK.

Andrew Cowie, DeepMind, London, UK.

Nicole Hobbs, DeepMind, London, UK.

Pushmeet Kohli, DeepMind, London, UK.

Gerard Kleywegt, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Ewan Birney, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

Demis Hassabis, DeepMind, London, UK.

Sameer Velankar, European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton, UK.

FUNDING

Funding for open access charge: DeepMind.

Conflict of interest statement . None declared.

REFERENCES

Associated Data

Data Availability Statement

All the AlphaFold predictions are publicly available through multiple data-access mechanisms. Coordinate files in PDB and mmCIF formats are available in TAR archives per proteome through FTP at ftp://ftp.ebi.ac.uk/pub/databases/alphafold . Meta-information and URLs to individual UniProt accessions are available via a public API endpoint. For example, https://alphafold.ebi.ac.uk/api/prediction/Q92793 provides all the information for UniProt accession Q92793 ( https://www.alphafold.ebi.ac.uk/entry/Q92793 ).