The DOI matches its Crossref record; 72 of the 72 references it cites were found in a registry or an index; 19 of the 29 sources its claims cite were checked for support.
Improved prediction of protein-protein interactions using AlphaFold2
Open accessPreprint — published as 10.1038/s41467-022-28865-w · Paper page
Abstract
Predicting the structure of interacting protein chains is a fundamental step towards understanding protein function. Unfortunately, no computational method can produce accurate structures of protein complexes. AlphaFold2, has shown unprecedented levels of accuracy in modelling single chain protein structures. Here, we apply AlphaFold2 for the prediction of heterodimeric protein complexes. We find that the AlphaFold2 protocol together with optimized multiple sequence alignments, generate models with acceptable quality (DockQ≥0.23) for 63% of the dimers. From the predicted interfaces we create a simple function to predict the DockQ score which distinguishes acceptable from incorrect models as well as interacting from non-interacting proteins with state-of-art accuracy. We find that, using the predicted DockQ scores, we can identify 51% of all interacting pairs at 1% FPR. The protocol can be found at: https://gitlab.com/ElofssonLab/FoldDock .
Overview
CoChat Summary
TL;DR
AlphaFold2, combined with optimized paired MSAs, can fold-and-dock many heterodimeric protein complexes and predict model quality from the resulting interfaces. The derived pDockQ score separates acceptable from incorrect models and interacting from non-interacting proteins with state-of-art accuracy.
Key findings
- On the test set, AF2 + paired MSAs reached a success rate of 57.8% (856/1481), rising to 61.7% when five seeds were ranked by pDockQ; block diagonalization + paired MSAs with top-ranked models reached 62.7%. The best development-set result was 39.4% for AF2 + paired MSAs.
- pDockQ, a sigmoidal fit to interface pLDDT × log(interface contacts), predicted DockQ with an average error of 0.11 and separated acceptable models (DockQ ≥ 0.23) from incorrect ones with AUC 0.95; 31% of models were acceptable at 1% error and 54% at 10% error.
- For interacting vs non-interacting proteins (1481 positives, 5694 negatives), pDockQ gave AUC 0.87 and identified 51% of interacting pairs at 1% FPR, compared with 27%, 18%, and 13% for interface pLDDT, interface contacts, and interface residues, respectively.
- Performance varied by group: bacteria 73.7%, archaea 84.5%, eukarya 61%, virus 60%; E. coli 76.4%, S. cerevisiae 66.2%, H. sapiens 58.1%; helix interfaces 62%, sheet 59%, loop 53%; largest interface tertile 74% vs smallest 47%. AF2 outperformed tested docking methods, while AF-multimer reached 72.2% but was trained on the same data as the test set.
Methods
- AF2 was used to model heterodimeric complexes with a chain break, no templates, and relaxation off. The development set contained 216 Dockground complexes; the test set contained 1481 complexes; additional sets included six CASP14 heterodimers, eight novel PDB complexes, and 5694 non-interacting pairs (3989 E. coli + 1705 mammalian).
- MSA strategies compared default AF2 MSAs, organism-paired MSAs, block diagonalization MSAs, and combinations; AF2 model_1 vs model_1_ptm and different recycle/ensemble configurations were tested, with five seeds for the best strategy.
- Models were assessed with DockQ (acceptable ≥ 0.23), ROC/AUC, and pDockQ; comparisons included RoseTTAFold, GRAMM, MDockPP, TMdock, and AF-multimer.
Limitations
- Stated: only heterodimeric states are considered; alternative oligomeric states and larger biological assemblies are outside scope. Stated: interaction/non-interaction separation needs extension to compare interactions within individual organisms. Stated: the development vs test success-rate discrepancy (39.4% vs 57.8%) suggests unknown selection bias, and the development set is much smaller (216 vs 1481).
- Evident: AF-multimer comparison is difficult because it was trained using the same data as the test set. Evident: the initial non-interacting set was only E. coli while positives were mainly eukaryotic, so the authors added mammalian negatives to obtain a more realistic estimate.
- Evident: several analyses excluded complexes due to memory limits, MSA errors, missing Cβs, or DSSP failures (e.g., final test set 1481; DSSP 1391; RoseTTAFold 1455; AF-multimer 1458), which may affect generalizability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures






Verification
Claims the paper makes
6 Valid · 14 Not valid · 5 Not checked · 76 Own results and interpretations
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“We find that the MSA generation process can be sped up substantially at no performance loss (performance increase of 1% SR) by simply fusing MSAs from two HHblits34 runs on Uniclust3035 instead of using the MSAs from AF2.”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Findings and future prospects
What the sources say
- Supportedcites [34] source
The citation is attached to “HHblits,” while “We find” presents the speed and success-rate comparison as the citing authors’ own result. The source’s Software section describes using HHblits to build query MSAs, supporting that method citation. (LLM, conf=0.88)
- Not checked — no full text of the sourcecites [35] source
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“The higher performance in prokaryotes is consistent with previous observations regarding the availability of evolutionary information in prokaryotes compared to Eukarya27 (Supplementary Fig. 1).”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Bacterial complexes are predicted more accurately
What the sources say
- Could not verifycites [27] sourceExperimental — accuracy not yet measured
The Results section on PNN removal and the Discussion address grid-cell activity in rats, not evolutionary information or performance comparisons between prokaryotes and Eukarya. Searches for those terms and the claim’s distinctive phrase found no passage addressing the claim in our copy. (LLM, conf=0.78)
plainstates an associationn = 1,481- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“The study of AF2s ability to separate interacting and non-interacting proteins here contains more extensive data than recent studies27.”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Limitations
What the sources say
- Could not verifycites [27] sourceExperimental — accuracy not yet measured
The Results section, “Reduced pairwise grid cell stability in a novel environment” (page unavailable), discusses correlations between rat grid cells after perineuronal-net degradation. Searches for AF2 and for interacting versus non-interacting proteins found no passage addressing that topic or comparing the extent of data with recent studies. (LLM, conf=0.78)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Another recently published method obtains AUC 0.76 on this set27.”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Identifying interacting proteins
What the sources say
- Could not verifycites [27] sourceExperimental — accuracy not yet measured
The Results section reports 0.76 as a median spatial correlation for control grid cells, not as an AUC. A search for “AUC 0.76” and “area under the curve” found no passage reporting an AUC for a method on the cited set; the paper’s reported analyses concern grid-cell activity. (LLM, conf=0.76)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“There are currently 64,006 pairwise human protein interactions in the human reference interactome36.”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Findings and future prospects
What the sources say
- Could not verifycites [36] sourceExperimental — accuracy not yet measured
The Results section reports counts for the E. coli interactome, not a human reference interactome. Searches for “64,006 pairwise human protein interactions” and “human reference interactome” found no passage in our copy giving the claimed human interaction count. (LLM, conf=0.76)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
34 more claims · 3 with their cited sources checked
Create a free account to see all 101 claims and every source verdict.
- It may state two separate claims as one. 38
- Probably not a substantial claim. 36
- The quote only partly states this claim. 1
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“A much better solution is to modify the “residue_index” feature used by AF2, which eliminates the need for a covalent linker that likely creates artifacts20,21.”
From AF2Complex predicts direct physical interactions in multimeric proteins with deep learning (2022) · cites this paper as [21]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“A large fraction of all interacting pairs have high ipTM/ranking confidence scores, whereas this is extremely rare for two randomly chosen proteins (3).”
From Reliable Identification of Homodimers Using AlphaFold (2025) · cites this paper as [3]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Model quality and interface confidence were assessed using the interfacial iPTM scores, pDockQ2 score calculated based on ipSAE [160-162].”
From Global identification of neuronal and astrocytic integral membrane proteins that require Retromer for their endosomal recycling (2026) · cites this paper as [161]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Bryant et al showed that AlphaFold2.0 could be used to successfully separate interacting and non-interacting heterodimers (3).”
From Reliable Identification of Homodimers Using AlphaFold (2025) · cites this paper as [3]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Test set performance
“The average error overall is 0.14 DockQ score.”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Distinguishing acceptable from incorrect models
“This enables the prediction of the DockQ scores (pDockQ) in a continuous manner with an overall average error of 0.11 on the test set.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Improved prediction of protein-protein interactions using AlphaFold2 > Methods > AlphaFold2
“We rank the five models for each complex by the number of residues in the interface, giving the best result.”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Results and discussion > Model variation and ranking for the test set
“Although this is unachievable, ranking the models using the pDockQ score results in an SR of 61.7%.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“Some structures in this dataset are homodimers (65) and are therefore excluded, resulting in 1705 structures.”
§ Improved prediction of protein-protein interactions using AlphaFold2 > Methods > Non-interacting proteins
Reported 1705 · recomputed 1899
Citations
Cites
Highly accurate protein structure prediction with AlphaFold
Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., et al. (2021) · Nature · doi:10.1038/s41586-021-03819-2 · Cited by 43,764 · cited 6×
“We note that performing model relaxation did not increase performance in the AF2 paper 15 and was, therefore, ignored to save computational cost.”
BackgroundMethodInfluentialSciPy 1.0: fundamental algorithms for scientific computing in Python
Virtanen, P., Gommers, R., Oliphant, T. E., Haberland, M., Reddy, T., Cournapeau, D., Burovski, E., Peterson, P., Weckesser, W., Bright, J., et al. (2020) · Nature Methods · doi:10.1038/s41592-019-0686-2 · Cited by 38,430
Kabsch, W., & Sander, C. (1983) · Biopolymers · doi:10.1002/bip.360221211 · Cited by 15,320
SciPy 1.0: fundamental algorithms for scientific computing in Python
Virtanen, P., Gommers, R., Oliphant, T. E., Haberland, M., Reddy, T., Cournapeau, D., Burovski, E., Peterson, P., Weckesser, W., Bright, J., et al. (2019) · Apollo - University of Cambridge Repository · doi:10.17863/cam.64134 · Cited by 10,460
Jalview Version 2—a multiple sequence alignment editor and analysis workbench
Waterhouse, A. M., Procter, J. B., Martin, D. M. A., Clamp, M., & Barton, G. J. (2009) · Bioinformatics · doi:10.1093/bioinformatics/btp033 · Cited by 10,235
Szklarczyk, D., Gable, A. L., Nastou, K. C., Lyon, D., Kirsch, R., Pyysalo, S., Doncheva, N. T., Legeay, M., Fang, T., Bork, P., et al. (2021) · Nucleic Acids Research · doi:10.1093/nar/gkaa1074 · Cited by 8,056
Accelerated Profile HMM Searches
Eddy, S. R. (2011) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1002195 · Cited by 6,975
UniProt: the universal protein knowledgebase in 2021
The UniProt Consortium, Bateman, A., Martin, M.-J., Orchard, S., Magrane, M., Agivetova, R., Ahmad, S., Alpi, E., Bowler-Barnett, E. H., Britto, R., et al. (2021) · Nucleic Acids Research · doi:10.1093/nar/gkaa1100 · Cited by 6,372
Accurate prediction of protein structures and interactions using a 3-track network
Baek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee, G. R., Wang, J., Cong, Q., Kinch, L. N., Schaeffer, R. D., et al. (2021) · doi:10.1101/2021.06.14.448402 · Cited by 5,549
MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets
Steinegger, M., & Söding, J. (2017) · Nature Biotechnology · doi:10.1038/nbt.3988 · Cited by 4,840
Cited by
Accurate structure prediction of biomolecular interactions with AlphaFold 3
Abramson, J., Adler, J., Dunger, J., Evans, R., Green, T., Pritzel, A., Ronneberger, O., Willmore, L., Ballard, A. J., Bambrick, J., et al. (2024) · Nature · doi:10.1038/s41586-024-07487-w · Cited by 15,861
ColabFold - Making protein folding accessible to all
Mirdita, M., Schütze, K., Moriwaki, Y., Heo, L., Ovchinnikov, S., & Steinegger, M. (2021) · doi:10.1101/2021.08.15.456425 · Cited by 9,391
“22 ).”
MethodProtein complex prediction with AlphaFold-Multimer
Evans, R., O’Neill, M., Pritzel, A., Antropova, N., Senior, A., Green, T., Žídek, A., Bates, R., Blackwell, S., Yim, J., et al. (2021) · doi:10.1101/2021.10.04.463034 · Cited by 3,808
Harnessing protein folding neural networks for peptide–protein docking
Tsaban, T., Varga, J. K., Avraham, O., Ben-Aharon, Z., Khramushin, A., & Schueler-Furman, O. (2022) · Nature Communications · doi:10.1038/s41467-021-27838-9 · Cited by 1,205
A structural biology community assessment of AlphaFold2 applications
Akdel, M., Pires, D. E. V., Pardo, E. P., Jänes, J., Zalevsky, A. O., Mészáros, B., Bryant, P., Good, L. L., Laskowski, R. A., Pozzati, G., et al. (2022) · Nature Structural & Molecular Biology · doi:10.1038/s41594-022-00849-w · Cited by 675
“To evaluate performance, one model for each pair was generated with AF2 (using default parameters, except that model_2 was used, providing a complementary set of results to those derived in (Bryant et al., 2021)).”
MethodSelf-citation
Metadata
- Authors
- Bryant, P., Pozzati, G., & Elofsson, A.
- Affiliations
- Stockholm University (SE)
- Corresponding author
- Bryant, P.
- Elofsson, A.
- Venue
- Nature Communications
- Publisher
- openRxiv
- Published
- 2021-09-15
- Volume
- 13
- Issue
- 1
- Pages
- 1265-1265
- ISSN
- 2041-1723 (ISSN-L)
- PMC
- PMC8913741
- OpenAlex
- W4220824576
- Cited by
- 1,022
- References
- 66
- Citations by year
- 2026: 170 · 2025: 269 · 2024: 292 · 2023: 203 · 2022: 83 · 2021: 5
- FWCI
- 62.72
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- Locations
- Nature Communications (publishedVersion, cc-by)
- Research Square (submittedVersion, cc-by)
- bioRxiv (Cold Spring Harbor Laboratory) (submittedVersion, 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
- Swedish e-Science Research Centre
- Knut och Alice Wallenbergs Stiftelse (COVID 19)
- Vetenskapsrådet (2016-06301, Berzelius-2021-29)
- Stockholms Universitet
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Structural Biology › Enzyme Structure and Function
- Keywords
- protein-protein interaction, protein complex structure prediction, AlphaFold2, multiple sequence alignment, protein interface prediction
- MeSH terms
- Protein Conformation, Proteins, Computational Biology
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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: Improved prediction of protein-protein interactions using AlphaFold2. https://app.cochat.ai/papers/10.1101/2021.09.15.460468
- The paper
- Bryant, P., Pozzati, G., & Elofsson, A. (2021). Improved prediction of protein-protein interactions using AlphaFold2. Nature Communications. https://doi.org/10.1101/2021.09.15.460468
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.