Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 106 of the 106 references it cites were found in a registry or an index.

Harnessing protein folding neural networks for peptide–protein docking

Open access
Tsaban, Tomer, Varga, Julia K., Avraham, Orly, Ben-Aharon, Ziv, Khramushin, Alisa, Schueler-Furman, Ora · 2022 · Nature Communications

Published version of a preprint

Abstract

Highly accurate protein structure predictions by deep neural networks such as AlphaFold2 and RoseTTAFold have tremendous impact on structural biology and beyond. Here, we show that, although these deep learning approaches have originally been developed for the in silico folding of protein monomers, AlphaFold2 also enables quick and accurate modeling of peptide–protein interactions. Our simple implementation of AlphaFold2 generates peptide–protein complex models without requiring multiple sequence alignment information for the peptide partner, and can handle binding-induced conformational changes of the receptor. We explore what AlphaFold2 has memorized and learned, and describe specific examples that highlight differences compared to state-of-the-art peptide docking protocol PIPER-FlexPepDock. These results show that AlphaFold2 holds great promise for providing structural insight into a wide range of peptide–protein complexes, serving as a starting point for the detailed characterization and manipulation of these interactions.

Overview

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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

Integrity signals

Registry

Crossref · 10.1038/s41467-021-27838-9

Cites

106 of 106 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 AF2 can be adapted to accurately model many peptide–protein interactions. a Concept of peptide–protein docking with a poly-glycine linker, successfully identified as unstructured region and modeled as a circle, placing the peptide at its correct position (left; example PDB ID 1ssh63), or out in space (right; example PDB ID 2orz64). The native peptide is shown in black. b Cumulative performance of AF2 and PFPD (dashed lines) on motif (12 complexes, blue), non-motif (14 complexes, red), and LNR (96 complexes, AF2 only, yellow) sets, as measured over the interface residues of the peptide (Left - Peptide interface: after aligning the receptor) or the full interface (Right - CAPRI Irms: after aligning the whole interface. See also Supplementary Fig. 2). c Correlation between performance of AF2 and PFPD for the motif and non-motif PFPD sets. Triangles indicate values over 15.0 Å RMSD. Left: PDB ID 2b9h38, interaction between MAPK Fus3 and a peptide derived from MAPKK Ste7, where PFPD positions the peptide within the pocket, but in a flipped orientation (N-termini are indicated by spheres). Right: PDB ID 1awr65, interaction between CypA and a HIV-1 Gag polyprotein derived peptide, where AF2 models the peptide within the pocket but in a wrong position. Shown are the peptide structures generated by the linker model (cyan, blue) or PFPD (magenta), and the crystal structure (black). d Overall assessment of performance measured for the individual partners (after aligning each separately). Source data are provided as a Source Data file.
Fig. 1 AF2 can be adapted to accurately model many peptide–protein interactions. a Concept of peptide–protein docking with a poly-glycine linker, successfully identified as unstructured region and modeled as a circle, placing the peptide at its correct position (left; example PDB ID 1ssh63), or out in space (right; example PDB ID 2orz64). The native peptide is shown in black. b Cumulative performance of AF2 and PFPD (dashed lines) on motif (12 complexes, blue), non-motif (14 complexes, red), and LNR (96 complexes, AF2 only, yellow) sets, as measured over the interface residues of the peptide (Left - Peptide interface: after aligning the receptor) or the full interface (Right - CAPRI Irms: after aligning the whole interface. See also Supplementary Fig. 2). c Correlation between performance of AF2 and PFPD for the motif and non-motif PFPD sets. Triangles indicate values over 15.0 Å RMSD. Left: PDB ID 2b9h38, interaction between MAPK Fus3 and a peptide derived from MAPKK Ste7, where PFPD positions the peptide within the pocket, but in a flipped orientation (N-termini are indicated by spheres). Right: PDB ID 1awr65, interaction between CypA and a HIV-1 Gag polyprotein derived peptide, where AF2 models the peptide within the pocket but in a wrong position. Shown are the peptide structures generated by the linker model (cyan, blue) or PFPD (magenta), and the crystal structure (black). d Overall assessment of performance measured for the individual partners (after aligning each separately). Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 2 Accurate modeling results in identification of motif residues and correlates with high pLDDT values. a Heatmaps of per residue RMSD (shades of green) for peptide–protein interactions with known motifs (motif set), with the top bar highlighting the motif region in dark blue, followed by corresponding pLDDT heatmaps (shades of brown). Each row corresponds to the network parameters used to generate the model (1 through 5). b Heatmaps for the non-motif set. Note 2b9h, for which reinspection of the literature reveals a motif. c Scatterplots of pLDDT vs. RMSD values for the different datasets, representing peptide per residue values of the AF2 models. For the LNR set, clusters of points are represented by hexagonal bins, colored in shades of yellow according to the number of residues in the cluster. The number of observations in each subsection is noted therein. The motif set is shown in shades of blue and the non-motif set in shades of red. Models with linker are represented by dark colored “x” and models with no linker (separate chains) are represented with pale “+”. Source data are provided as a Source Data file.
Fig. 2 Accurate modeling results in identification of motif residues and correlates with high pLDDT values. a Heatmaps of per residue RMSD (shades of green) for peptide–protein interactions with known motifs (motif set), with the top bar highlighting the motif region in dark blue, followed by corresponding pLDDT heatmaps (shades of brown). Each row corresponds to the network parameters used to generate the model (1 through 5). b Heatmaps for the non-motif set. Note 2b9h, for which reinspection of the literature reveals a motif. c Scatterplots of pLDDT vs. RMSD values for the different datasets, representing peptide per residue values of the AF2 models. For the LNR set, clusters of points are represented by hexagonal bins, colored in shades of yellow according to the number of residues in the cluster. The number of observations in each subsection is noted therein. The motif set is shown in shades of blue and the non-motif set in shades of red. Models with linker are represented by dark colored “x” and models with no linker (separate chains) are represented with pale “+”. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 3 Recovery of peptide–protein interface characteristics by AF2 models. a AF2 models identify a significant fraction of the receptor binding pocket residues, even though coverage is reduced compared to PFPD models. Box-and-whisker plots are shown for motif (n = 12, blue) and non-motif (n = 14, red) sets, with median as center line, quartiles as box limits, and lower/upper whiskers extending to the maximum/minimum data points within the interquartile range, respectively. b Computational alanine scanning recovers well-predicted interface hotspots, mostly for accurate models (within 2.5 Å RMSD): Comparison of computational Rosetta alanine scanning results applied to the best model vs. the native structure. The vertical and horizontal lines represent the threshold ΔΔG = 1.5 kcal/mol used to define interface hotspots. The different colors represent distinct bins of RMSD (of the model used for alanine scanning). The number of residues in each quadrant is indicated (counts from models below 2.5 Å are shown in parentheses). c Performance is significantly reduced if the peptide sequence is changed to poly-alanine, both for motif and non-motif sets. Source data are provided as a Source Data file.
Fig. 3 Recovery of peptide–protein interface characteristics by AF2 models. a AF2 models identify a significant fraction of the receptor binding pocket residues, even though coverage is reduced compared to PFPD models. Box-and-whisker plots are shown for motif (n = 12, blue) and non-motif (n = 14, red) sets, with median as center line, quartiles as box limits, and lower/upper whiskers extending to the maximum/minimum data points within the interquartile range, respectively. b Computational alanine scanning recovers well-predicted interface hotspots, mostly for accurate models (within 2.5 Å RMSD): Comparison of computational Rosetta alanine scanning results applied to the best model vs. the native structure. The vertical and horizontal lines represent the threshold ΔΔG = 1.5 kcal/mol used to define interface hotspots. The different colors represent distinct bins of RMSD (of the model used for alanine scanning). The number of residues in each quadrant is indicated (counts from models below 2.5 Å are shown in parentheses). c Performance is significantly reduced if the peptide sequence is changed to poly-alanine, both for motif and non-motif sets. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 4 AF2 modeling of interactions involving significant changes in receptor conformation. a Estrogen receptor alpha - peptide derived from nuclear receptor cofactor 2 (PDB ID 2b1z66). Left: In the free receptor conformation (PDB ID 3ert67) the C-terminal helix occupies the peptide binding site. Shown are the solved crystal structures of the unbound (blue) and bound conformation (receptor in white, peptide in green). Right: AF2 models the correct receptor conformation, and positions the peptide accurately in the binding site (RMSD < 1.0 Å). Shown are the AF2 model (unbound receptor: maroon, bound receptor: pale pink, peptide: magenta), compared to bound conformation (white). b EphB4 receptor––ephrin-B2 antagonist (PDB ID 2bba68). Left: In the free receptor conformation (PDB ID 3etp69) the J–K loop forms a β-hairpin. Right: Upon peptide binding, this hairpin becomes disordered and assumes a flexible loop conformation, which is recovered in the AF2 model. Coloring as in a.
Fig. 4 AF2 modeling of interactions involving significant changes in receptor conformation. a Estrogen receptor alpha - peptide derived from nuclear receptor cofactor 2 (PDB ID 2b1z66). Left: In the free receptor conformation (PDB ID 3ert67) the C-terminal helix occupies the peptide binding site. Shown are the solved crystal structures of the unbound (blue) and bound conformation (receptor in white, peptide in green). Right: AF2 models the correct receptor conformation, and positions the peptide accurately in the binding site (RMSD < 1.0 Å). Shown are the AF2 model (unbound receptor: maroon, bound receptor: pale pink, peptide: magenta), compared to bound conformation (white). b EphB4 receptor––ephrin-B2 antagonist (PDB ID 2bba68). Left: In the free receptor conformation (PDB ID 3etp69) the J–K loop forms a β-hairpin. Right: Upon peptide binding, this hairpin becomes disordered and assumes a flexible loop conformation, which is recovered in the AF2 model. Coloring as in a.CC BY · PubMed Central
Fig. 5 Analysis of potential factors influencing AF2 peptide docking performance. a Peptide length does not affect docking performance: successful models are generated for long as well as short peptides (See Supplementary Fig. 8A for the distribution of peptide lengths in the LNR dataset). Values over 20 Å are indicated by triangles. b Alpha helical peptides are modeled particularly well. Cumulative plot of performance according to the secondary structure of the peptide. (See Supplementary Fig. 8b for the corresponding distribution of secondary structures). Helical, beta strand, and coiled peptides are colored in pink, blue and gray, respectively. c Examples of potential memorization by AF2 from similar monomer structures: Left: Highly accurate prediction where a precise coverage of the structure of the peptide-receptor interface is available (native PDB ID 1ssc70 and memorized PDB ID 1a5p71). Center: Inaccurate template but successful modeling (native PDB ID 3ayu72 and memorized PDB ID 5ue473). Right: Inaccurate template and failed model (native PDB ID 2x7274 and memorized PDB ID 5dgy75). In the latter two cases the model is most probably not built based on the inaccurate template. For all panels, the native receptor is shown in white, and native peptide in black, the memorized peptide in blue, and the modeled peptide in green. d Cumulative plot of performance on a set of peptide-receptor complex structures that were excluded due to post-translational modifications or bound ligands that influence the peptide structure: AF2 modeling is not dramatically impaired. The LNR set is shown in yellow, and the PTM + LIG set is shown in black. Source data are provided as a Source Data file.
Fig. 5 Analysis of potential factors influencing AF2 peptide docking performance. a Peptide length does not affect docking performance: successful models are generated for long as well as short peptides (See Supplementary Fig. 8A for the distribution of peptide lengths in the LNR dataset). Values over 20 Å are indicated by triangles. b Alpha helical peptides are modeled particularly well. Cumulative plot of performance according to the secondary structure of the peptide. (See Supplementary Fig. 8b for the corresponding distribution of secondary structures). Helical, beta strand, and coiled peptides are colored in pink, blue and gray, respectively. c Examples of potential memorization by AF2 from similar monomer structures: Left: Highly accurate prediction where a precise coverage of the structure of the peptide-receptor interface is available (native PDB ID 1ssc70 and memorized PDB ID 1a5p71). Center: Inaccurate template but successful modeling (native PDB ID 3ayu72 and memorized PDB ID 5ue473). Right: Inaccurate template and failed model (native PDB ID 2x7274 and memorized PDB ID 5dgy75). In the latter two cases the model is most probably not built based on the inaccurate template. For all panels, the native receptor is shown in white, and native peptide in black, the memorized peptide in blue, and the modeled peptide in green. d Cumulative plot of performance on a set of peptide-receptor complex structures that were excluded due to post-translational modifications or bound ligands that influence the peptide structure: AF2 modeling is not dramatically impaired. The LNR set is shown in yellow, and the PTM + LIG set is shown in black. Source data are provided as a Source Data file.CC BY · PubMed Central

Verification

Claims the paper makes

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

3 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

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 103 references

Cited by

See all 1,205

Metadata

Authors
Tsaban, T., Varga, J. K., Avraham, O., Ben-Aharon, Z., Khramushin, A., & Schueler-Furman, O.
Affiliations
Hebrew University of Jerusalem (IL)
Venue
Nature Communications
Published
2022-01-10
Volume
13
Issue
1
Pages
176-176
ISSN
2041-1723 (ISSN-L)
OpenAlex
W4220887281
Cited by
1,205
References
103
Citations by year
2026: 146 · 2025: 262 · 2024: 303 · 2023: 329 · 2022: 153
FWCI
85.05
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • Research Square (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Israel Science Foundation (717/2017, 301/2021)
  • United States-Israel Binational Science Foundation (207/2015)
  • European Commission (860517 — UBIMOTIF)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › vaccines and immunoinformatics approaches
Keywords
protein-peptide docking, AlphaFold2, protein folding, protein-peptide complex, deep learning protein structure prediction, multiple sequence alignment, protein structure modeling
MeSH terms
Protein Conformation, alpha-Helical, Protein Conformation, beta-Strand, Amino Acid Sequence, Binding Sites, Models, Molecular, Peptides, Protein Binding, Proteins, Software, Neural Networks, Computer, Protein Folding, Protein Interaction Mapping, Protein Interaction Domains and Motifs, Molecular Docking Simulation

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Harnessing protein folding neural networks for peptide–protein docking. https://app.cochat.ai/papers/10.1038/s41467-021-27838-9
The paper
Tsaban, T., Varga, J. K., Avraham, O., Ben-Aharon, Z., Khramushin, A., & Schueler-Furman, O. (2022). Harnessing protein folding neural networks for peptide–protein docking. Nature Communications. https://doi.org/10.1038/s41467-021-27838-9

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.