AlphaFold predictions of fold-switched conformations are driven by structure memorization
Open accessPublished version of a preprint
Abstract
Recent work suggests that AlphaFold (AF)–a deep learning-based model that can accurately infer protein structure from sequence–may discern important features of folded protein energy landscapes, defined by the diversity and frequency of different conformations in the folded state. Here, we test the limits of its predictive power on fold-switching proteins, which assume two structures with regions of distinct secondary and/or tertiary structure. We find that (1) AF is a weak predictor of fold switching and (2) some of its successes result from memorization of training-set structures rather than learned protein energetics. Combining >280,000 models from several implementations of AF2 and AF3, a 35% success rate was achieved for fold switchers likely in AF’s training sets. AF2’s confidence metrics selected against models consistent with experimentally determined fold-switching structures and failed to discriminate between low and high energy conformations. Further, AF captured only one out of seven experimentally confirmed fold switchers outside of its training sets despite extensive sampling of an additional ~280,000 models. Several observations indicate that AF2 has memorized structural information during training, and AF3 misassigns coevolutionary restraints. These limitations constrain the scope of successful predictions, highlighting the need for physically based methods that readily predict multiple protein conformations.
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.
Integrity signals
Registry
Cites
Statistics
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.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
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 · see citing works
Matplotlib: a 2D graphics environment
Hunter, J. D. (2007) · Computing in Science & Engineering · doi:10.1109/mcse.2007.55 · Cited by 39,249 · see citing works
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 · see citing works
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 · see citing works
seaborn: statistical data visualization
Waskom, M. (2021) · Journal of Open Source Software · doi:10.21105/joss.03021 · Cited by 7,310 · see citing works
Protein 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 · see citing works
Improved prediction of protein-protein interactions using AlphaFold2
Bryant, P., Pozzati, G., & Elofsson, A. (2021) · doi:10.1101/2021.09.15.460468 · Cited by 1,025 · see citing works
Predicting multiple conformations via sequence clustering and AlphaFold2
Wayment-Steele, H. K., Ojoawo, A., Otten, R., Apitz, J. M., Pitsawong, W., Hömberger, M., Ovchinnikov, S., Colwell, L., & Kern, D. (2024) · Nature · doi:10.1038/s41586-023-06832-9 · Cited by 545 · cited 10× · see citing works
“Since then, several recently developed enhanced sampling techniques have challenged our hypothesis, proposing instead that AF2 couples coevolution with a learned energy function to predict alternative conformations 10,15,17 .”
InfluentialAhdritz, G., Bouatta, N., Floristean, C., Kadyan, S., Xia, Q., Gerecke, W., O’Donnell, T. J., Berenberg, D., Fisk, I., Zanichelli, N., et al. (2024) · Nature Methods · doi:10.1038/s41592-024-02272-z · Cited by 439 · cited 2× · see citing works
“…contacts medium gray; coevolutionary information inferred from MSA using ACE, (teal). c Both AF2 multimer and AF2 predict the correctXCL1 dimerstructurefrom single sequences and 0 recycles, suggesting that they memorized its structure during training. match AF2 ’ s predictive performance 27 .”
InfluentialSPEACH_AF: Sampling protein ensembles and conformational heterogeneity with Alphafold2
Stein, R. A., & Mchaourab, H. S. (2022) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1010483 · Cited by 258 · see citing works
Impact of protein conformational diversity on AlphaFold predictions
Saldaño, T., Escobedo, N., Marchetti, J., Zea, D. J., Mac Donagh, J., Velez Rueda, A. J., Gonik, E., Melani, A. G., Novomisky Nechcoff, J., Salas, M. N., et al. (2021) · doi:10.1101/2021.10.27.466189 · Cited by 196 · cited 6× · see citing works
“De fi ning “ Complex ” and “ Easy ” fold-switchers Since, the 92 fold-switching protein pairs are highly heterogeneous with respect to the structural differences between conformers, a fair comparisonofthepredictionsproducedbyAF2samplingswouldneed a separation of the targets based on their dif fi culty or complexity.”
InfluentialHigh-throughput prediction of protein conformational distributions with subsampled AlphaFold2
Monteiro da Silva, G., Cui, J. Y., Dalgarno, D. C., Lisi, G. P., & Rubenstein, B. M. (2024) · Nature Communications · doi:10.1038/s41467-024-46715-9 · Cited by 196 · see citing works
AlphaFold, Artificial Intelligence (AI), and Allostery
Nussinov, R., Zhang, M., Liu, Y., & Jang, H. (2022) · The Journal of Physical Chemistry B · doi:10.1021/acs.jpcb.2c04346 · Cited by 190 · see citing works
Current structure predictors are not learning the physics of protein folding
Outeiral, C., Nissley, D. A., & Deane, C. M. (2022) · Bioinformatics · doi:10.1093/bioinformatics/btab881 · Cited by 82 · see citing works
Breaking the conformational ensemble barrier: Ensemble structure modeling challenges in CASP15
Kryshtafovych, A., Montelione, G., Rigden, D., Mesdaghi, S., Karaca, E., & Moult, J. (2023) · doi:10.22541/au.169199504.48580490/v1 · Cited by 31 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
LassoPred: a tool to predict the 3D structure of lasso peptides
Ouyang, X., Ran, X., Xu, H., Al-Abssi, R., Zhao, Y.-L., Link, A. J., & Yang, Z. J. (2025) · Nature Communications · doi:10.1038/s41467-025-60544-4 · Cited by 11 · see citing works
Lidbrink, S. E., Howard, R. J., Haloi, N., & Lindahl, E. (2025) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1013187 · Cited by 9 · see citing works
Wang, B., Wang, C., Chen, J., Liu, D., Sun, C., Zhang, J., Zhang, K., & Li, H. (2026) · Nature Machine Intelligence · doi:10.1038/s42256-026-01198-9 · Cited by 5 · see citing works
- Authors
- Chakravarty, D., Schafer, J. W., Chen, E. A., Thole, J. F., Ronish, L. A., Lee, M., & Porter, L. L.
- Affiliations
- National Institutes of Health (US)
- Corresponding author
- Porter, Lauren L.
- ORCID
- Venue
- Nature Communications
- Published
- 2024-08-24
- Volume
- 15
- Issue
- 1
- Pages
- 7296-7296
- ISSN
- 2041-1723 (ISSN-L)
- PMC
- PMC11344769
- OpenAlex
- W4401844424
- Cited by
- 159
- References
- 15
- Citations by year
- 2026: 57 · 2025: 77 · 2024: 25
- FWCI
- 21.58
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- Nature Communications (publishedVersion, cc-by)
- bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (submittedVersion, public-domain)
- RePEc: Research Papers in Economics (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- Funders
- U.S. Department of Health & Human Services | NIH | U.S. National Library of Medicine (LM202011)
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
- Physical Sciences › Materials Science › Materials Chemistry › Enzyme Structure and Function
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
- Keywords
- AlphaFold, protein conformational switching, protein energy landscape, protein structure prediction, AlphaFold2, AlphaFold3
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
