CoChat
Sign Up
PreprintCited by 2,598Protein Structure and DynamicsEnzyme Structure and FunctionMachine Learning in Bioinformatics

The DOI matches its Crossref record; 46 of the 48 references it cites were found in a registry or an index.

Fast and accurate protein structure search with Foldseek

Open access
van Kempen, Michel, Kim, Stephanie S., Tumescheit, Charlotte, Mirdita, Milot, Lee, Jeongjae, Gilchrist, Cameron L.M., Söding, Johannes, Steinegger, Martin·2022·Nature Biotechnology
DOI 10.1101/2022.02.07.479398PMC PMC10869269

Preprint — published as 10.1038/s41587-023-01773-0 · Paper page

Open this paper in CoChat — free accountOpen PDF

Abstract

As structure prediction methods are generating millions of publicly available protein structures, searching these databases is becoming a bottleneck. Foldseek aligns the structure of a query protein against a database by describing the amino acid backbone of proteins as sequences over a structural alphabet. Foldseek decreases computation times by four to five orders of magnitude with 86%, 88% and 133% of the sensitivities of DALI, TM-align and CE, respectively.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1101/2022.02.07.479398

Cites

46 of 46 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Foldseek workflow. a, Foldseek searches a set of query structures through a set of target structures. (1) Query and target structures are discretized into 3Di sequences (see b). To detect candidate structures, we apply the fast and sensitive k-mer and ungapped alignment prefilter of MMseqs2 to the 3Di sequences, (2) followed by vectorized Smith–Waterman local alignment combining 3Di and amino acid substitution scores. Alternatively, a global alignment is computed with a 1.7-times accelerated TM-align version (Supplementary Fig. 12). b, Learning the 3Di alphabet. (1) 3Di states describe tertiary interaction between a residue i and its nearest neighbor j. Nearest neighbors have the closest virtual center distance (yellow). Virtual center positions (Supplementary Fig. 1) were optimized for maximum search sensitivity. (2) To describe the interaction geometry of residues i and j, we extract seven angles, the Euclidean Cα distance and two sequence distance features from the six Cα coordinates of the two backbone fragments (blue and red). (3) These 10 features are used to define 20 3Di states by training a VQ-VAE28 modified to learn states that are maximally evolutionary conserved. For structure searches, the encoder predicts the best-matching 3Di state for each residue.
Fig. 1 Foldseek workflow. a, Foldseek searches a set of query structures through a set of target structures. (1) Query and target structures are discretized into 3Di sequences (see b). To detect candidate structures, we apply the fast and sensitive k-mer and ungapped alignment prefilter of MMseqs2 to the 3Di sequences, (2) followed by vectorized Smith–Waterman local alignment combining 3Di and amino acid substitution scores. Alternatively, a global alignment is computed with a 1.7-times accelerated TM-align version (Supplementary Fig. 12). b, Learning the 3Di alphabet. (1) 3Di states describe tertiary interaction between a residue i and its nearest neighbor j. Nearest neighbors have the closest virtual center distance (yellow). Virtual center positions (Supplementary Fig. 1) were optimized for maximum search sensitivity. (2) To describe the interaction geometry of residues i and j, we extract seven angles, the Euclidean Cα distance and two sequence distance features from the six Cα coordinates of the two backbone fragments (blue and red). (3) These 10 features are used to define 20 3Di states by training a VQ-VAE28 modified to learn states that are maximally evolutionary conserved. For structure searches, the encoder predicts the best-matching 3Di state for each residue.CC BY · PubMed Central
Fig. 2 Foldseek reaches similar sensitivities as structural aligners at thousands of times their speed. a, Cumulative distributions of sensitivity for homology detection on the SCOPe40 database of single-domain structures. TPs are matches within the same superfamily; FPs are matches between different folds. Sensitivity is the area under the ROC (AUROC) curve up to the first FP (see Supplementary Fig. 4 for family and fold). b, Precision-recall curve of SCOPe40 superfamilies (see Supplementary Fig. 4 for family and fold). c, Average sensitivity up to the first FP for family, superfamily and fold versus total runtime on an AMD EPYC 7702P 64-core CPU for the all-versus-all searches of 11,211 structures of SCOPe40. d, Search sensitivity on multi-domain, full-length AlphaFold2 protein models. One hundred queries, randomly selected from AlphaFoldDB (version 1), were searched against this database. Per-residue query coverage (y axis) is the fraction of residues covered by at least x (x axis) TP matches ranked before the first FP match. e, Alignment quality for alignments of AlphaFoldDB (version 1) protein models (top panel), averaged over the top five matches of each of the 100 queries. Sensitivity = TP residues in alignment / query length; precision = TP residues / alignment length. Reference-based alignment quality benchmark on HOMSTRAD alignments. f, Alignment quality comparison between Foldseek and Dali for each HOMSTRAD family. The F1 score is the harmonic mean between sensitivity and precision.
Fig. 2 Foldseek reaches similar sensitivities as structural aligners at thousands of times their speed. a, Cumulative distributions of sensitivity for homology detection on the SCOPe40 database of single-domain structures. TPs are matches within the same superfamily; FPs are matches between different folds. Sensitivity is the area under the ROC (AUROC) curve up to the first FP (see Supplementary Fig. 4 for family and fold). b, Precision-recall curve of SCOPe40 superfamilies (see Supplementary Fig. 4 for family and fold). c, Average sensitivity up to the first FP for family, superfamily and fold versus total runtime on an AMD EPYC 7702P 64-core CPU for the all-versus-all searches of 11,211 structures of SCOPe40. d, Search sensitivity on multi-domain, full-length AlphaFold2 protein models. One hundred queries, randomly selected from AlphaFoldDB (version 1), were searched against this database. Per-residue query coverage (y axis) is the fraction of residues covered by at least x (x axis) TP matches ranked before the first FP match. e, Alignment quality for alignments of AlphaFoldDB (version 1) protein models (top panel), averaged over the top five matches of each of the 100 queries. Sensitivity = TP residues in alignment / query length; precision = TP residues / alignment length. Reference-based alignment quality benchmark on HOMSTRAD alignments. f, Alignment quality comparison between Foldseek and Dali for each HOMSTRAD family. The F1 score is the harmonic mean between sensitivity and precision.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 48 references

Research works printed but not linked: 16

  1. 10.1126/science.abj8754 · DOI · Paper page
  2. Lin, Z. et al. bioRxiv (2022). 2022.07.20.500902.
  3. 10.1038/nbt.3988 · DOI · Paper page
  4. Kim, H. et al. bioRxiv 2022.12.09.519715 (2022).
  5. 2009, Journal of Statistical Software, 32, 1
  6. Van den Oord, A. et al. Adv Neur Inf Proc Syst (NIPS) 30 (2017). · DOI
  7. 10.1093/NAR/GKAB1061 · DOI · Paper page
  8. 10.1093/bioinformatics/btq078 · DOI
  9. 10.1006/jmbi.1990.9999 · DOI · Paper page
  10. 10.1038/S41586-021-03819-2 · DOI · Paper page
  11. 10.1093/bioinformatics/btt473 · DOI · Paper page
  12. 10.1093/protein/11.9.739 · DOI
  13. 2020, Bioinformatics, 36, i718, 10.1093/bioinformatics/btaa839 · DOI
  14. 10.1002/pro.5560071126 · DOI
  15. Lin, Z. et al. bioRxiv (2022). 2022.07.20.500902.
  16. 10.1093/nar/gki524 · DOI · Paper page

Cited by

Sorted by citations

Related

See all

Metadata

Authors
van Kempen, M., Kim, S. S., Tumescheit, C., Mirdita, M., Lee, J., Gilchrist, C. L. M., Söding, J., & Steinegger, M.
Affiliations
  • Max Planck Institute for Multidisciplinary Sciences (DE)
  • Seoul National University (KR)
Corresponding author
  • Söding, Johannes
  • Steinegger, Martin
ORCID
Venue
Nature Biotechnology
Publisher
openRxiv
Published
2022-02-09
Volume
42
Issue
2
Pages
243-246
ISSN
0733-222X (ISSN-L)
DOI
10.1101/2022.02.07.479398
PMC
PMC10869269
OpenAlex
W4375858802
Cited by
2,598
References
48
Citations by year
2026: 830 · 2025: 917 · 2024: 592 · 2023: 211 · 2022: 44
FWCI
226.70
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Nature Biotechnology (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • MPG.PuRe (Max Planck Society) (submittedVersion, other-oa)
  • GoeScholar The Publication Server of the Georg-August-Universität Göttingen (Georg-August-Universität Göttingen) (submittedVersion, cc-by)
  • Seoul National University Open Repository (Seoul National University) (submittedVersion)
Funders
  • National Research Foundation
  • Bundesministerium für Bildung und Forschung
  • Seoul National University
  • National Research Foundation of Korea (2021-R1C1-C102065, 2019R1A6A1A10073437, 2020M3A9G7103933, 2021R1C1C102065, 2021M3A9I4021220, 2021-M3A9-I4021220, 2020M3‐A9G7‐103933, 2019R1A6A1, 2019R1-A6A1-A10073437)
  • Max-Planck-Gesellschaft
  • Samsung
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Structural Biology › Enzyme Structure and Function
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
Keywords
protein structure search, Foldseek, structural alignment, structural alphabet, protein structure database, computational efficiency
MeSH terms
Algorithms, Amino Acids, Proteins, Software, Databases, Protein

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review