AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination
Open accessPublished version of a preprint
Abstract
Artificial intelligence-based protein structure prediction methods such as AlphaFold have revolutionized structural biology. The accuracies of these predictions vary, however, and they do not take into account ligands, covalent modifications or other environmental factors. Here, we evaluate how well AlphaFold predictions can be expected to describe the structure of a protein by comparing predictions directly with experimental crystallographic maps. In many cases, AlphaFold predictions matched experimental maps remarkably closely. In other cases, even very high-confidence predictions differed from experimental maps on a global scale through distortion and domain orientation, and on a local scale in backbone and side-chain conformation. We suggest considering AlphaFold predictions as exceptionally useful hypotheses. We further suggest that it is important to consider the confidence in prediction when interpreting AlphaFold predictions and to carry out experimental structure determination to verify structural details, particularly those that involve interactions not included in the prediction.
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.
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?
Drug binding often depends on protein motion, such as small shifts when a compound binds, long-range effects within the protein, hidden pockets that open and close, or changes when proteins form complexes.
“Drug binding often depends on motion, such as small shifts when a compound binds, long-range effects within the protein, hidden pockets that open and close, or changes when proteins form complexes.”
From Artificial Intelligence in Clinical Oncology: From Productivity Enhancement to Creative Discovery (2025) · cites this paper as [106]
Cancer-related mutations and chemical modifications can alter protein shapes.
“Cancer-related mutations and chemical modifications can also alter these shapes.”
From Artificial Intelligence in Clinical Oncology: From Productivity Enhancement to Creative Discovery (2025) · cites this paper as [106]
AlphaFold predicts a protein's most stable shape but does not model how the protein moves.
“It predicts a protein’s most stable shape; however, it does not model how the protein moves.”
From Artificial Intelligence in Clinical Oncology: From Productivity Enhancement to Creative Discovery (2025) · cites this paper as [106]
AlphaFold has limitations for drug discovery, including cancer therapeutics.
“AlphaFold has been a breakthrough for drug discovery, including cancer therapeutics, but its limitations are well recognized [105,106,107,108].”
From Artificial Intelligence in Clinical Oncology: From Productivity Enhancement to Creative Discovery (2025) · cites this paper as [106]
Human researchers still needed to test and refine AlphaFold's output to ensure that the proposed structure is 100% accurate.
“Although human researchers still needed to test and refine AlphaFold’s output to ensure that the proposed structure is 100% accurate, the ML system greatly improves the efficiency of protein chemistry research [216].”
From The ethics of using artificial intelligence in scientific research: new guidance needed for a new tool (2025) · cites this paper as [216]
The ML system AlphaFold greatly improves the efficiency of protein chemistry research.
“Although human researchers still needed to test and refine AlphaFold’s output to ensure that the proposed structure is 100% accurate, the ML system greatly improves the efficiency of protein chemistry research [216].”
From The ethics of using artificial intelligence in scientific research: new guidance needed for a new tool (2025) · cites this paper as [216]
Terwilliger et al estimate about 10% of residues AF2 predicts with high confidence deviate by more than 2 Å from native backbone conformations.
“Terwilliger et al estimate that ~10% of residues predicted by AF2 with high confidence deviate from the backbone by more than 2 Å from native conformations observed in experimental structures61”
From The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins (2024) · cites this paper as [61]
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
Evolutionary-scale prediction of atomic level protein structure with a language model
Lin, Z., Akin, H., Rao, R., Hie, B., Zhu, Z., Lu, W., Smetanin, N., Verkuil, R., Kabeli, O., Shmueli, Y., et al. (2022) · doi:10.1101/2022.07.20.500902 · Cited by 5,523 · see citing works
Towards automated crystallographic structure refinement with <i>phenix.refine</i>
Afonine, P. V., Grosse-Kunstleve, R. W., Echols, N., Headd, J. J., Moriarty, N. W., Mustyakimov, M., Terwilliger, T. C., Urzhumtsev, A., Zwart, P. H., & Adams, P. D. (2012) · Acta Crystallographica Section D Biological Crystallography · doi:10.1107/s0907444912001308 · Cited by 5,511 · see citing works
Self-citationProtein 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
Highly accurate protein structure prediction for the human proteome
Tunyasuvunakool, K., Adler, J., Wu, Z., Green, T., Zielinski, M., Žídek, A., Bridgland, A., Cowie, A., Meyer, C., Laydon, A., et al. (2021) · Nature · doi:10.1038/s41586-021-03828-1 · Cited by 3,025 · cited 2× · see citing works
“Of course, many of the proteins in the human proteome that have low-confidence AlphaFold predictions are likely to contain regions that are intrinsically disordered 24,25 that would also often not be revealed by experimental methods.”
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 · cited 1× · see citing works
“…predictions vary substantially in their global and local agreement with deposited models and also in their coverage at the highest levels of confidence 1,11,21 , with only 36% of residues in the human proteome 22 and 73% of residues in Escherichia coli modeled with very high confidence 23 .”
Self-citationSPEACH_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
AlphaFold heralds a data-driven revolution in biology and medicine
Thornton, J. M., Laskowski, R. A., & Borkakoti, N. (2021) · Nature Medicine · doi:10.1038/s41591-021-01533-0 · Cited by 185 · see citing works
The structural coverage of the human proteome before and after AlphaFold
Porta-Pardo, E., Ruiz-Serra, V., Valentini, S., & Valencia, A. (2022) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1009818 · Cited by 131 · see citing works
‘The entire protein universe’: AI predicts shape of nearly every known protein
Callaway, E. (2022) · Nature · doi:10.1038/d41586-022-02083-2 · Cited by 113 · cited 2× · see citing works
“…using AlphaFold 1 , RoseTTAFold 2 and related methods 3 are far more accurate than previous generations of prediction algorithms 4 , bringing much closer to reality the biological understanding that could be derived from knowing the three-dimensional structures of all macromolecules 1,2,5–9 .”
Barbarin-Bocahu, I., & Graille, M. (2022) · Acta Crystallographica Section D Structural Biology · doi:10.1107/s2059798322002157 · Cited by 52 · see citing works
Crystallographic molecular replacement using an in silico‐generated search model of SARS‐CoV‐2 ORF8
Flower, T. G., & Hurley, J. H. (2021) · Protein Science · doi:10.1002/pro.4050 · Cited by 49 · see citing works
Flagged first, then by citations
Research works printed but not linked: 28
- [2]Baek M Accurate prediction of protein structures and interactions using a three-track neural network Science 2021 373 871 876 10.1126/science.abj8754 34282049 PMC7612213 · DOI · Paper page
- [4]Kryshtafovych, A., Schwede, T., Topf, M., Fidelis, K. & Moult, J. Critical assessment of methods of protein structure prediction (CASP)—round XIV. Proteins89, 1607–1617 (2021). PMC8726744 34533838
- [7]van Breugel M Rosa e Silva I Andreeva A Structural validation and assessment of AlphaFold2 predictions for centrosomal and centriolar proteins and their complexes Commun. Biol. 2022 5 312 10.1038/s42003-022-03269-0 35383272 PMC8983713 · DOI
- [8]Subramaniam S Kleywegt GJ A paradigm shift in structural biology Nat. Methods 2022 19 20 23 10.1038/s41592-021-01361-7 35017736 · DOI
- [9]Ourmazd A Moffat K Lattman EE Structural biology is solved—now what? Nat. Methods 2022 19 24 26 10.1038/s41592-021-01357-3 35017723 · DOI
- [11]Shao C Bittrich S Wang S Burley SK Assessing PDB macromolecular crystal structure confidence at the individual amino acid residue level Structure 2022 30 1385 1394 10.1016/j.str.2022.08.004 36049478 PMC9547844 · DOI
- [12]Goulet A Cambillau C Present impact of AlphaFold2 revolution on structural biology, and an illustration with the structure prediction of the bacteriophage J-1 host adhesion device Front. Mol. Biosci. 2022 9 907452 10.3389/fmolb.2022.907452 35615740 PMC9124777 · DOI
- [13]Moore PB Hendrickson WA Henderson R Brunger AT The protein-folding problem: not yet solved Science 2022 375 507 10.1126/science.abn9422 35113705 · DOI
- [14]Acharya KR Lloyd MD The advantages and limitations of protein crystal structures Trends Pharmacol. Sci. 2005 26 10 14 10.1016/j.tips.2004.10.011 15629199 · DOI
- [15]Fraser JS Accessing protein conformational ensembles using room-temperature X-ray crystallography Proc. Natl Acad. Sci. USA 2011 108 16247 16252 10.1073/pnas.1111325108 21918110 PMC3182744 · DOI
- [18]wwPDB Consortium. Protein Data Bank: the single global archive for 3D macromolecular structure data Nucleic Acids Res. 2018 47 D520 D528 10.1093/nar/gky949 PMC6324056 30357364 · DOI · Paper page
- [19]Jumper J Hassabis D Protein structure predictions to atomic accuracy with AlphaFold Nat. Methods 2022 19 11 12 10.1038/s41592-021-01362-6 35017726 · DOI · Paper page
- [20]van Beusekom B Joosten K Hekkelman ML Joosten RP Perrakis A Homology-based loop modeling yields more complete crystallographic protein structures IUCrJ 2018 5 585 594 10.1107/S2052252518010552 30224962 PMC6126648 · DOI
- [21]Hryc CF Baker ML AlphaFold2 and CryoEM: revisiting CryoEM modeling in near-atomic resolution density maps iScience 2022 25 104496 10.1016/j.isci.2022.104496 35733789 PMC9207676 · DOI
- [24]Dunker AK Intrinsically disordered protein J. Mol. Graph. Model. 2001 19 26 59 10.1016/S1093-3263(00)00138-8 11381529 · DOI · Paper page
- [27]Terwilliger TC Improved AlphaFold modeling with implicit experimental information Nat. Methods 2022 19 1376 1382 10.1038/s41592-022-01645-6 36266465 PMC9636017 · DOI · Paper page
- [28]Zhang Q Re-sensitization of mcr carrying multidrug resistant bacteria to colistin by silver Proc. Natl Acad. Sci. USA 2022 119 e2119417119 10.1073/pnas.2119417119 35263219 PMC8931383 · DOI
- [29]Burkhardt I de Rond T Chen PY-T Moore BS Ancient plant-like terpene biosynthesis in corals Nat. Chem. Biol. 2022 18 664 669 10.1038/s41589-022-01026-2 35606558 PMC9179088 · DOI
- [30]Hobbs SJ Phage anti-CBASS and anti-Pycsar nucleases subvert bacterial immunity Nature 2022 605 522 526 10.1038/s41586-022-04716-y 35395152 PMC9117128 · DOI
- [31]Terwilliger, T. C. et al. Accelerating crystal structure determination with iterative AlphaFold prediction. Acta Crystallogr.D42, 234–244 (2023). PMC9986801 36876433
- [33]Bartlett GJ Porter CT Borkakoti N Thornton JM Analysis of catalytic residues in enzyme active sites J. Mol. Biol. 2002 324 105 121 10.1016/S0022-2836(02)01036-7 12421562 · DOI
- [34]McCoy AJ Sammito MD Read RJ Implications of AlphaFold2 for crystallographic phasing by molecular replacement Acta Crystallogr. D Struct. Biol. 2022 78 1 13 10.1107/S2059798321012122 34981757 PMC8725160 · DOI · Paper page
- [36]Oeffner RD Putting AlphaFold models to work with phenix.process_predicted_model and ISOLDE Acta Crystallogr. D Struct. Biol. 2022 78 1303 1314 10.1107/S2059798322010026 36322415 PMC9629492 · DOI · Paper page
- [37]Terwilliger T Maximum-likelihood density modification Acta Crystallogr. D Biol. Crystalogr. 2000 56 965 972 10.1107/S0907444900005072 PMC2792768 10944333 · DOI
- [38]Cowtan K Metcalfe S Bond P Shift-field refinement of macromolecular atomic models Acta Crystallogr. D Struct. Biol. 2020 76 1192 1200 10.1107/S2059798320013170 33263325 PMC7709196 · DOI
- [40]Pettersen EF UCSF ChimeraX: structure visualization for researchers, educators, and developers Protein Sci. 2021 30 70 82 10.1002/pro.3943 32881101 PMC7737788 · DOI
- [41]Davis IW Arendall WB Richardson DC Richardson JS The backrub motion: how protein backbone shrugs when a sidechain dances Structure 2006 14 265 274 10.1016/j.str.2005.10.007 16472746 · DOI
- [42]Bennett MJ Choe S Eisenberg D Domain swapping: entangling alliances between proteins Proc. Natl Acad. Sci. USA 1994 91 3127 3131 10.1073/pnas.91.8.3127 8159715 PMC43528 · DOI
Cited by
Sorted by citations
Resnik, D. B., & Hosseini, M. (2025) · AI and Ethics · doi:10.1007/s43681-024-00493-8 · Cited by 228 · see citing works
“Although human researchers still needed to test and refine AlphaFold’s output to ensure that the proposed structure is 100% accurate, the ML system greatly improves the efficiency of protein chemistry research [216].”
The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins
Agarwal, V., & McShan, A. C. (2024) · Nature Chemical Biology · doi:10.1038/s41589-024-01638-w · Cited by 155 · see citing works
“Terwilliger et al. 61 estimate that ~10% of residues predicted by AF2 with high confidence deviate from the backbone by more than 2 Å from native conformations observed in experimental structures.”
BackgroundChallenges in bridging the gap between protein structure prediction and functional interpretation
Varadi, M., Tsenkov, M., & Velankar, S. (2025) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.26614 · Cited by 36 · see citing works
Modeling the Orthosteric Binding Site of the G Protein-Coupled Odorant Receptor OR5K1
Nicoli, A., Haag, F., Marcinek, P., He, R., Kreißl, J., Stein, J., Marchetto, A., Dunkel, A., Hofmann, T., Krautwurst, D., et al. (2023) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.2c00752 · Cited by 28 · see citing works
Wu, D., Yin, R., Chen, G., Ribeiro-Filho, H. V., Cheung, M., Robbins, P. F., Mariuzza, R. A., & Pierce, B. G. (2024) · Science Advances · doi:10.1126/sciadv.adq6150 · Cited by 27 · see citing works
Zhang, Y., Zhang, Z., Kagaya, Y., Terashi, G., Zhao, B., Xiong, Y., & Kihara, D. (2025) · Communications Biology · doi:10.1038/s42003-025-08783-5 · Cited by 19 · see citing works
“…cases, predicting a different domain orientation from the corresponding Protein Data Bank (PDB) 20 entry is not necessarily incorrect; however, it becomes a problem when the predicted model shows inconsistencies with experimental density data 21 , Nuclear Magnetic Resonance (NMR), or cryo-EM maps.”
Rational design of 19F NMR labelling sites to probe protein structure and interactions
Streit, J. O., Chan, S. H. S., Daya, S., & Christodoulou, J. (2025) · Nature Communications · doi:10.1038/s41467-025-59105-6 · Cited by 17 · see citing works
Probing the modulation of enzyme kinetics by multi-temperature, time-resolved serial crystallography
Schulz, E. C., Prester, A., von Stetten, D., Gore, G., Hatton, C. E., Bartels, K., Leimkohl, J.-P., Schikora, H., Ginn, H. M., Tellkamp, F., et al. (2025) · Nature Communications · doi:10.1038/s41467-025-61631-2 · Cited by 15 · see citing works
High-throughput Selection of Human <i>de novo</i> -emerged sORFs with High Folding Potential
Aubel, M., Buchel, F., Heames, B., Jones, A., Honc, O., Bornberg-Bauer, E., & Hlouchova, K. (2024) · Genome Biology and Evolution · doi:10.1093/gbe/evae069 · Cited by 12 · see citing works
Severe deviation in protein fold prediction by advanced AI: a case study
López-Sagaseta, J., & Urdiciain, A. (2025) · Scientific Reports · doi:10.1038/s41598-025-89516-w · Cited by 10 · see citing works
Artificial Intelligence in Clinical Oncology: From Productivity Enhancement to Creative Discovery
Kuno, M., Osumi, H., Udagawa, S., Yoshikawa, K., Ooki, A., Shinozaki, E., Ishikawa, T., Oba, J., Yamaguchi, K., & Sakurada, K. (2025) · Current Oncology · doi:10.3390/curroncol32110588 · Cited by 7 · see citing works
Unifying Constraints Linking Protein Folding and Native Dynamics Decoded from AlphaFold
Zhang, Z., Ren, W., Xie, L., Zheng, Y., Guan, X., Wang, J., Li, W., & Tang, Q.-Y. (2026) · Physical Review Letters · doi:10.1103/7j2j-f8f7 · Cited by 5 · 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
- Terwilliger, T. C., Liebschner, D., Croll, T. I., Williams, C. J., McCoy, A. J., Poon, B. K., Afonine, P. V., Oeffner, R. D., Richardson, J. S., Read, R. J., & Adams, P. D.
- Affiliations
- Los Alamos National Laboratory (US)
- Lawrence Berkeley National Laboratory (US)
- University of Cambridge (GB)
- Duke University (US)
- Corresponding author
- Terwilliger, Thomas C.
- ORCID
- Venue
- Nature Methods
- Published
- 2023-11-30
- Volume
- 21
- Issue
- 1
- Pages
- 110-116
- ISSN
- 1548-7091 (ISSN-L)
- 1548-7105
- PMC
- PMC10776388
- OpenAlex
- W4389174567
- Cited by
- 411
- References
- 12
- Citations by year
- 2026: 123 · 2025: 150 · 2024: 113 · 2023: 22 · 2022: 2 · 2013: 1
- FWCI
- 36.04
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- Nature Methods (publishedVersion, cc-by)
- bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- eScholarship (California Digital Library) (submittedVersion, cc-by)
- Apollo (University of Cambridge) (submittedVersion, cc-by)
- OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion)
- Apollo (University of Cambridge) (submittedVersion)
- Apollo
- Funders
- U.S. Department of Health & Human Services | NIH | National Institute of General Medical Sciences (GM063210)
- Wellcome Trust (209407/Z/17/Z)
- DOE | LDRD | Lawrence Berkeley National Laboratory (DE-AC02-05CH11231)
- 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 structure prediction, prediction accuracy, prediction confidence, protein conformation, side-chain conformation, domain orientation, hypothesis generation
- MeSH terms
- Artificial Intelligence, Crystallography, Mental Processes, Protein Conformation
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
