The DOI matches its Crossref record; 75 of the 77 references it cites were found in a registry or an index; 4 of the 13 sources its claims cite were checked for support.
Accurate structure prediction of biomolecular interactions with AlphaFold 3
Open accessAbstract
The introduction of AlphaFold 2 1 has spurred a revolution in modelling the structure of proteins and their interactions, enabling a huge range of applications in protein modelling and design 2–6 . Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues. The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer v.2.3 7,8 . Together, these results show that high-accuracy modelling across biomolecular space is possible within a single unified deep-learning framework.
Overview
CoChat Summary
TL;DR
AlphaFold 3 (AF3) is a unified diffusion-based deep-learning model that predicts joint structures of complexes containing proteins, nucleic acids, small molecules, ions and modified residues. It reports substantially improved accuracy over specialized tools across protein–ligand, protein–nucleic acid, protein–protein and antibody–antigen tasks.
Key findings
- AF3 outperforms docking tools on protein–ligand interfaces: on the 428-structure PoseBusters benchmark, it beats Vina (Fisher’s exact test, P = 2.27 × 10−13) and RoseTTAFold All-Atom (P = 4.45 × 10−25), using pocket-aligned ligand r.m.s.d. < 2 Å as the success metric.
- AF3 improves protein–protein and antibody–antigen prediction over AlphaFold-Multimer v.2.3: protein–protein success (DockQ > 0.23) increases (paired Wilcoxon signed-rank test, P = 1.8 × 10−18), antibody–protein prediction improves markedly (P = 6.5 × 10−5), and protein monomer LDDT improves (P = 1.7 × 10−34).
- AF3 predicts protein–nucleic acid complexes and RNA with higher accuracy than RoseTTAFold2NA, and on ten CASP15 RNA targets achieves higher average performance than RoseTTAFold2NA and AIchemy_RNA, but does not reach the best human-expert-aided CASP15 submission AIchemy_RNA2.
- Covalent modifications are also predicted: success rates (pocket r.m.s.d. < 2 Å) are 42.1% for multi-residue glycans (n = 131 clusters) and 46.1% for single-residue glycans on all-quality experimental data (n = 167). Example predictions include PDB 7PZB (full-complex LDDT 82.8; GDT 90.1) and PDB 7PNM (LDDT 83.0; GDT 83.1).
Methods
- AF3 replaces the AF2 evoformer with a simpler pairformer module (48 blocks) and replaces the AF2 structure module with a diffusion module that directly predicts raw atom coordinates without rotational frames or equivariant processing; MSA processing is substantially de-emphasized.
- Training used no structural data released after 30 September 2021; PoseBusters evaluation used a model with a 30 September 2019 cut-off. Training had three stages with crop sizes of 384, 640 and 768 tokens; one optimizer step used a mini batch of 256 samples and 12,288 diffusion samples initially, reduced to 8,192 during fine-tuning.
- Evaluation used the PoseBusters set (428 protein–ligand structures released 2021 or later), a recent PDB set of 8,856 complexes released between 1 May 2022 and 12 January 2023, and ten publicly available CASP15 RNA targets. Metrics included DockQ, LDDT, pocket-aligned r.m.s.d., iLDDT, pLDDT, PAE and PDE.
Limitations
- The paper states stereochemical limitations: a chirality violation rate of 4.4% on the PoseBusters benchmark despite a ranking penalty, and occasional clashing atoms, with almost all remaining clashes occurring for protein–nucleic complexes with both greater than 100 nucleotides and greater than 2,000 residues in total.
- The paper states that the diffusion model can hallucinate spurious structural order in disordered regions, that AF3 predicts static PDB-like structures rather than solution dynamics, and that modelled conformational states may be incorrect or incomplete, such as predicting only the closed cereblon state for both apo and holo systems.
- The paper states that antibody–antigen accuracy keeps improving with more model seeds even up to 1,000 (Wilcoxon signed-rank test between 5 and 1,000 seeds, P = 2.0 × 10−5 for percentage correct and P = 0.009 for percentage very high accuracy), incurring extra computational cost; it also notes that CASP15 RNA significance statistics are not reported owing to limited dataset sizes.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures














Verification
Claims the paper makes
3 Valid · 2 Not valid · 10 Not checked · 48 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Similarly to some recent work34, we find that no invariance or equivariance with respect to global rotations and translation of the molecule are required in the architecture and we therefore omit them to simplify the machine learning architecture.”
§ Accurate structure prediction of biomolecular interactions with AlphaFold 3 > Network architecture and training
What the sources say
- Supportedcites [34] Wang, Y., Elhag, A. A., Jaitly, N., Susskind, J. M. & Bautista, M. A. Generating molecular conformer fields. Preprint at arXiv https://doi.org/10.48550/arXiv.2311.17932 (2023). source
Section 3.3 (p. 4) says that explicitly enforcing roto-translation equivariance is not a strong requirement for generalization in graph-conditioned conformer generation, and that the authors’ model does not enforce it. This supports the citing sentence’s architectural point; the paper distinguishes this setting from conformer likelihood models. (LLM, conf=0.91)
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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“A class of targets in which we observe this effect strongly is antibody–antigen complexes, similar to other recent work46.”
- Also stated in: § Model limitations · plain
“Figure 5a shows that, for AF3, top-ranked predictions keep improving with more model seeds, even at as many as 1,000 (Wilcoxon signed-rank test between 5 and 1,000 seeds, P = 2.0 × 10−5 for percentage correct and P = 0.009 for percentage very high accuracy; ranking by protein–protein interface ipTM).”
§ Accurate structure prediction of biomolecular interactions with AlphaFold 3 > Model limitations
What the sources say
- Partly supportedcites [46] Wallner, B. AFsample: improving multimer prediction with AlphaFold using massive sampling. Bioinformatics 39, btad573 (2023). sourceExperimental — accuracy not yet measured
Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.
boostedstates 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.
- Also stated in: § Model limitations · plain
Every source the paper cites for this claim was checked and supports it.
In the paper
“Generally, protein–protein prediction success (DockQ > 0.23)40 has increased (paired Wilcoxon signed-rank test, P = 1.8 × 10−18)”
§ Accurate structure prediction of biomolecular interactions with AlphaFold 3 > Accuracy across complex types
What the sources say
- Supportedcites [40] Basu, S. & Wallner, B. DockQ: a quality measure for protein-protein docking models. PLoS ONE 11, e0161879 (2016). source
The citation is attached to the DockQ > 0.23 threshold, not to the reported increase or Wilcoxon test. In “Optimizing d1 and d2” (page unavailable), the source identifies 0.23 as the cutoff separating Incorrect from Acceptable docking models. (LLM, conf=0.93)
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
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This cross-distillation greatly reduced the hallucination behaviour of AF3 (Extended Data Fig. 1 for disorder prediction results on the CAID 236 benchmark set).”
§ Accurate structure prediction of biomolecular interactions with AlphaFold 3 > Network architecture and training
boostedstates 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.
- Rests on an assumption not shown here.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AF3 can predict structures from input polymer sequences, residue modifications and ligand SMILES (simplified molecular-input line-entry system).”
§ Accurate structure prediction of biomolecular interactions with AlphaFold 3 > Accuracy across complex types
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.
34 more claims
Create a free account to see all 63 claims and to verify the 1 reference still open.
- It may state two separate claims as one. 13
- Probably not a substantial claim. 13
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
“Increasingly, deep learning models are used to predict and control complex systems we do not yet fully understand, such as inertially confined plasmas (Degrave et al., 2022) and biomoloecular systems (Abramson et al., 2024).”
From Robust agents learn causal world models (2024) · cites this paper as Abramson et al., 2024
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Structural modeling50,51 of the Fli1-Spi1 and Cebpa-Gata2 interactions suggested that steric hinderance cannot explain these effects (Figures 4F and S3H–S3K; see STAR Methods, section structural prediction of TF-TF-DNA complexes).”
From Design principles of cell-state-specific enhancers in hematopoiesis (2025) · cites this paper as [50]
The cited source says what the claim attributes to it.
In the citing paper
“While the capability of these models to predict structures from intrinsically disordered regions is limited,50 together with our functional data on spacing dependency, these results suggest that specific protein-protein interactions between these TFs or their co-factors may play a role in the switch from activators to repressors.”
From Design principles of cell-state-specific enhancers in hematopoiesis (2025) · cites this paper as [50]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“The AlphaFold3 cofolded structure prediction for EvoCas9–1 has a high mean pLDDT score of 90 across its protein, RNA, and DNA components”
From Sequence modeling and design from molecular to genome scale with Evo (2024) · cites this paper as [64]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Although various frameworks already exist for tackling specific biological questions (for example, protein structure prediction models89 were developed in the context of the CASP evaluation framework), the AIVC will need to demonstrate generalizability across numerous biological contexts and downstream tasks.”
From How to build the virtual cell with artificial intelligence: Priorities and opportunities (2024) · cites this paper as [89]
Experimental — accuracy not yet measured
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 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
§ Accurate structure prediction of biomolecular interactions with AlphaFold 3 > Accuracy across complex types
“We also evaluated AF3 performance on the ten publicly available Critical Assessment of Structure Prediction 15 (CASP15) RNA targets”
§ Accurate structure prediction of biomolecular interactions with AlphaFold 3 > Methods > Nucleic acid prediction baseline
“(R1116/8S95, R1117/8FZA, R1126 (downloaded from the CASP15 website https://predictioncenter.org/casp15/TARGETS_PDB/R1126.pdb), R1128/8BTZ, R1136/7ZJ4, R1138/[7PTK/7PTL], R1189/7YR7 and R1190/7YR6)”
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
Self-citationBerman, H. M. (2000) · Nucleic Acids Research · doi:10.1093/nar/28.1.235 · Cited by 40,196
Trott, O., & Olson, A. J. (2010) · Journal of Computational Chemistry · doi:10.1002/jcc.21334 · Cited by 38,412 · cited 2×
“Even so, AF3 greatly outperforms classical docking tools such as Vina 37,38 even while not using any structural inputs (Fisher’s exact test, P = 2.27 × 10 −13 ) and greatly outperforms all other true blind docking like RoseTTAFold All-Atom ( P = 4.45 × 10 −25 ).”
BackgroundInfluentialEvolutionary-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 5,990
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
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025 · cited 1×
“The biggest issue is that generative models are prone to hallucination 35 , whereby the model may invent plausible-looking structure even in unstructured regions.”
BackgroundProtein 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 · cited 3×
“…accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer v.2.3 7,8 .”
BackgroundMethodInfluentialSelf-citationScoring function for automated assessment of protein structure template quality
Zhang, Y., & Skolnick, J. (2004) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.20264 · Cited by 2,392
Impact of cytosine methylation on DNA binding specificities of human transcription factors
Yin, Y., Morgunova, E., Jolma, A., Kaasinen, E., Sahu, B., Khund-Sayeed, S., Das, P. K., Kivioja, T., Dave, K., Zhong, F., et al. (2017) · Science · doi:10.1126/science.aaj2239 · Cited by 1,257
Mariani, V., Biasini, M., Barbato, A., & Schwede, T. (2013) · Bioinformatics · doi:10.1093/bioinformatics/btt473 · Cited by 1,138
Research works printed but not linked: 1
- Stärk, H., Ganea, O., Pattanaik, L., Barzilay, D. & Jaakkola, T. EquiBind: Geometric deep learning for drug binding structure prediction. In Proc. 39th International Conference on Machine Learning (eds Chaudhuri, K. et al.) 20503–20521 (PMLR, 2022).
Cited by
Simulating 500 million years of evolution with a language model
Hayes, T., Rao, R., Akin, H., Sofroniew, N. J., Oktay, D., Lin, Z., Verkuil, R., Tran, V. Q., Deaton, J., Wiggert, M., et al. (2025) · Science · doi:10.1126/science.ads0018 · Cited by 937
Self-citationBoltz-2: Towards Accurate and Efficient Binding Affinity Prediction
Passaro, S., Corso, G., Wohlwend, J., Reveiz, M., Thaler, S., Somnath, V. R., Getz, N., Portnoi, T., Roy, J., Stark, H., et al. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.06.14.659707 · Cited by 665
Sequence modeling and design from molecular to genome scale with Evo
Nguyen, E., Poli, M., Durrant, M. G., Kang, B., Katrekar, D., Li, D. B., Bartie, L. J., Thomas, A. W., King, S. H., Brixi, G., et al. (2024) · Science · doi:10.1126/science.ado9336 · Cited by 580
“Visualizations of ISEvo1 TnpA and TnpB were also computed using AlphaFold3 by uploading sequences to the AlphaFold Server ( 64 ).”
Cancer treatments: Past, present, and future
Sonkin, D., Thomas, A., & Teicher, B. A. (2024) · Cancer Genetics · doi:10.1016/j.cancergen.2024.06.002 · Cited by 548
Chai-1: Decoding the molecular interactions of life
Discovery, C., Boitreaud, J., Dent, J., McPartlon, M., Meier, J., Reis, V., Rogozhnikov, A., & Wu, K. (2024) · doi:10.1101/2024.10.10.615955 · Cited by 517
Metadata
- Authors
- Abramson, J., Adler, J., Dunger, J., Evans, R., Green, T., Pritzel, A., Ronneberger, O., Willmore, L., Ballard, A. J., Bambrick, J., Bodenstein, S. W., Evans, D. A., Hung, C.-C., O’Neill, M., Reiman, D., Tunyasuvunakool, K., Wu, Z., Žemgulytė, A., Arvaniti, E., Beattie, C., Bertolli, O., Bridgland, A., Cherepanov, A., Congreve, M., Cowen-Rivers, A. I., Cowie, A., Figurnov, M., Fuchs, F. B., Gladman, H., Jain, R., Khan, Y. A., Low, C. M. R., Perlin, K., Potapenko, A., Savy, P., Singh, S., Stecula, A., Thillaisundaram, A., Tong, C., Yakneen, S., Zhong, E. D., Zielinski, M., Žídek, A., Bapst, V., Kohli, P., Jaderberg, M., Hassabis, D., & Jumper, J. M.
- Affiliations
- Google DeepMind (United Kingdom) (GB)
- Princeton University (US)
- Corresponding author
- Jaderberg, Max
- Hassabis, Demis
- Jumper, John M.
- ORCID
- Abramson, Josh
- Adler, Jonas
- Evans, Richard
- Green, Tim
- Pritzel, Alexander
- Ronneberger, Olaf
- Willmore, Lindsay
- Ballard, Andrew J.
- Bambrick, Joshua
- Bodenstein, Sebastian W.
- Hung, Chia-Chun
- O’Neill, Michael
- Reiman, David
- Tunyasuvunakool, Kathryn
- Wu, Zachary
- Arvaniti, Eirini
- Beattie, Charles
- Bertolli, Ottavia
- Cherepanov, Alexey
- Congreve, Miles
- Cowen-Rivers, Alexander I.
- Cowie, Andrew
- Figurnov, Michael
- Jain, Rishub
- Khan, Yousuf A.
- Low, Caroline M. R.
- Perlin, Kuba
- Potapenko, Anna
- Savy, Pascal
- Singh, Sukhdeep
- Stecula, Adrian
- Tong, Catherine
- Yakneen, Sergei
- Zhong, Ellen D.
- Žídek, Augustin
- Bapst, Victor
- Kohli, Pushmeet
- Jaderberg, Max
- Hassabis, Demis
- Jumper, John M.
- Venue
- Nature
- Published
- 2024-05-08
- Volume
- 630
- Issue
- 8016
- Pages
- 493-500
- ISSN
- 0028-0836 (ISSN-L)
- 1476-4687
- PMC
- PMC11168924
- OpenAlex
- W4396721167
- Semantic Scholar
- 7572ba7f604ef95d7acdd657ebac458106bd35df
- Cited by
- 15,861
- References
- 77
- Citations by year
- 2026: 7419 · 2025: 6885 · 2024: 1538
- FWCI
- 2149.54
- Type
- journal article
- Language
- en
- Update notices
- addendum (2024-11-27)
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Nature (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- 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 › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
- Keywords
- AlphaFold3, protein structure prediction, biomolecular interactions, protein-ligand interactions, protein-nucleic acid interactions, protein complexes, deep learning
- MeSH terms
- Deep Learning, Antibodies, Antigens, Humans, Ions, Ligands, Models, Molecular, Nucleic Acids, Protein Binding, Protein Conformation, Proteins, Software, Reproducibility of Results, 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.
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: Accurate structure prediction of biomolecular interactions with AlphaFold 3. https://app.cochat.ai/papers/10.1038/s41586-024-07487-w
- The paper
- Abramson, J., Adler, J., Dunger, J., Evans, R., Green, T., Pritzel, A., Ronneberger, O., Willmore, L., Ballard, A. J., Bambrick, J., Bodenstein, S. W., Evans, D. A., Hung, C.-C., O’Neill, M., Reiman, D., Tunyasuvunakool, K., Wu, Z., Žemgulytė, A., Arvaniti, E., . . . Jumper, J. M. (2024). Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature. https://doi.org/10.1038/s41586-024-07487-w
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.