The DOI matches its Crossref record; 38 of the 40 references it cites were found in a registry or an index; 4 of the 19 sources its claims cite were checked for support.
ColabFold - Making protein folding accessible to all
Open accessPreprint — published as 10.1038/s41592-022-01488-1 · Paper page
Abstract
ColabFold offers accelerated protein structure and complex predictions by combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold. ColabFold’s 40 - 60× faster search and optimized model use allows predicting close to a thousand structures per day on a server with one GPU. Coupled with Google Colaboratory, ColabFold becomes a free and accessible platform for protein folding. ColabFold is open-source software available at github.com/sokrypton/ColabFold . Its novel environmental databases are available at colabfold.mmseqs.com Contact: milot.mirdita@mpibpc.mpg.de , so@fas.harvard.edu , martin.steinegger@snu.ac.kr
Overview
CoChat Summary
TL;DR
ColabFold accelerates protein structure and complex prediction by replacing AlphaFold2/RoseTTAFold homology search with MMseqs2, making folding accessible via Google Colaboratory and local command-line tools. It reports 40–60-fold faster MSA generation, ~90-fold batch speed-up, and prediction accuracy matching AlphaFold2 on CASP14 and AlphaFold-multimer on ClusPro.
Key findings
- ColabFold’s MMseqs2 search is 40–60-fold faster than AlphaFold2’s homology search, and batch prediction is ~90-fold faster by avoiding recompilation and adding early stopping; this enables close to 1,000 structures per day on one GPU.
- On CASP14 free-modeling targets, mean TM-scores were 0.826 for ColabFold-AlphaFold2-BFD/MGnify, 0.818 for ColabFold-AlphaFold2-ColabFoldDB, 0.79 for AlphaFold2, 0.744 for AlphaFold-Colab, and 0.62 for ColabFold-RoseTTAFold-BFD/MGnify. Over all CASP14 targets, TM-scores were 0.887, 0.886, 0.888, and 0.754 for the respective methods excluding AlphaFold-Colab.
- ColabFold achieved the highest accuracy for complex prediction on the ClusPro dataset with the AlphaFold-multimer model, though some targets performed better using residue-index mode.
- The 1,762-protein *Methanocaldococcus jannaschii* proteome (proteins <1,000 amino acids) was predicted in 48 h on one Nvidia Titan RTX with early stopping at pLDDT ≥85; average pLDDTs for AlphaFold2 and ColabFold Stop ≥85 were 89.75 and 88.78 in a subsampled set of 50 proteins.
Methods
- ColabFold combines an MMseqs2-based homology search server with AlphaFold2 or RoseTTAFold; it searches UniRef100/UniRef30, PDB70, and either BFD/MGnify or ColabFoldDB, using a diversity-aware filter and optional MSA pairing for complexes.
- Benchmarks included 65 CASP14 targets containing 91 domains (20 free-modeling targets with 28 domains), 17 ClusPro complex targets evaluated with DockQ, and the *M. jannaschii* proteome; predictions were compared to experimental structures using TM-align and DockQ.
- Run-time measurements separated MSA generation and model inference, using specified CPU/GPU systems; ColabFold modes were executed without templates for benchmarking.
Limitations
- The paper states limitations of Google Colaboratory and provides the `colabfold_batch` command-line tool for tasks too large for Colab; it also notes AlphaFold-Colab cannot be used as a standalone and was excluded from the all-CASP14-target comparison.
- Evident from the text: ColabFold benchmarks were run without templates, so template-enabled performance is not directly benchmarked; free-modeling CASP14 targets were used to optimize MMseqs2 search parameters, which may make those results optimistic.
- Evident from the text: validation is limited to CASP14, ClusPro, and one proteome, so generalization to other target classes or larger proteomes is not established here.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures


Verification
Claims the paper makes
4 Valid · 4 Not valid · 12 Not checked · 51 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Prediction quality depends on the input MSA, however, often an MSA with only a few (~30) sufficiently diverse sequences is enough to produce high-quality predictions (see fig. 5a in ref. 1).”
§ ColabFold: making protein folding accessible to all > Main
What the sources say
- Supportedcites [1] source
The 'MSA depth and cross-chain contacts' section states directly that 'The model uses MSAs and the accuracy decreases substantially when the median alignment depth is less than around 30 sequences (see Fig. 5a for details)', with a threshold effect above ~100 sequences — supporting both that quality depends on the input MSA and that ~30 sequences suffices. The Methods ('Inputs and data sources') repeats it: 'the depth of the MSA is relatively unimportant until it approaches a threshold value of around 30 sequences when the MSA size effects become quite large.' The 'sufficiently diverse' qualifier matches the Neff-weighted effective-sequence definition of MSA depth given in the Fig. 5 legend and Methods ('Test set of recent PDB sequences'). (LLM, conf=0.90)
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
Every source the paper cites for this claim was checked and supports it.
In the paper
“Prediction quality depends on the input MSA, however, often an MSA with only a few (~30) sufficiently diverse sequences is enough to produce high-quality predictions (see fig. 5a in ref. 1).”
§ ColabFold: making protein folding accessible to all > Main
What the sources say
- Supportedcites [1] source
The “MSA depth and cross-chain contacts” section reports that accuracy decreases substantially below a median alignment depth of about 30 effective sequences (Fig. 5a). The Methods (“Inputs and data sources”) likewise describe about 30 as the threshold below which MSA-size effects become large. This supports the claim that an MSA of roughly that depth can often suffice for high-quality predictions. (LLM, conf=0.91)
plainstates a possible 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
Every source the paper cites for this claim was checked and supports it.
In the paper
“Mirdita et al.11 improved it to also support fast single-against-many searches.”
§ ColabFold: making protein folding accessible to all > Methods > Faster MSA generation with MMseqs2
What the sources say
- Supportedcites [11] source
The Methods section (2.1) says the authors reduced MMseqs2’s per-search runtime overhead from about 0.5 minutes to 0.05 seconds. The Results section reports testing the improved version with a single-sequence query against a database of 13.5 million sequences, supporting the paraphrase that it enables fast single-against-many searches. (LLM, conf=0.92)
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
“We show that ColabFold outperforms AlphaFold-Colab and matches AlphaFold2 on CASP14 targets and also matches AlphaFold-multimer on the ClusPro4,12 dataset in prediction quality.”
§ ColabFold: making protein folding accessible to all > Main
boosted- 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.
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
“We show that ColabFold outperforms AlphaFold-Colab and matches AlphaFold2 on CASP14 targets and also matches AlphaFold-multimer on the ClusPro4,12 dataset in prediction quality.”
§ ColabFold: making protein folding accessible to all > Main
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.
32 more claims
Create a free account to see all 71 claims and to verify the 15 references still open.
- Probably not a substantial claim. 18
- It may state two separate claims as one. 17
- Borderline: one of the tests for a substantial claim is unclear. 1
- The quote only partly states this claim. 2
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
“allow users to run a modified AF2 protocol on open access servers in minutes”
From The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins (2024) · cites this paper as [8]
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 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 [20]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“it has been shown recently that contact predictions can be used to simultaneously fold and dock proteins17, and early reports have indicated that AF2 can predict the structure of complexes18–20, which it was not initially trained to handle.”
From A structural biology community assessment of AlphaFold2 applications (2022) · cites this paper as [19]
The cited source says what the claim attributes to it.
In the citing paper
“However, AF2 run using ColabFold34 predicts the thermodynamically unfavoured FS state for KaiBTE (Fig. 1c (left)).”
From Predicting multiple conformations via sequence clustering and AlphaFold2 (2024) · cites this paper as [34]
The cited source says what the claim attributes to it.
In the citing paper
“However, AlphaFold and RoseTTAfold have recently demonstrated unprecedented accuracy in predicting structures of monomeric proteins (27,28,(61)(62)(63)(64)(65)(66) and complexes (35,62,67,68).”
From AI-based structure prediction empowers integrative structural analysis of human nuclear pores (2022) · cites this paper as [35]
7 more claims about this paper
Create a free account to see every claim other papers make about this one.
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
“The average pLDDTs of AlphaFold2 and ColabFold Stop ≥ 85 were 89.75 and 88.78 in a subsampled set of 50 proteins.”
“AlphaFold2 (reduced_dbs, v2.1.1), ColabFold (commit f5d0cec) default and ColabFold Stop ≥ 85 have an average pLDDT of 90.68, 90.22 and 89.33, respectively, for 50 randomly sampled proteins.”
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-citationMatplotlib: a 2D graphics environment
Hunter, J. D. (2007) · Computing in Science & Engineering · doi:10.1109/mcse.2007.55 · Cited by 39,249
Unicycler: Resolving bacterial genome assemblies from short and long sequencing reads
Wick, R. R., Judd, L. M., Gorrie, C. L., & Holt, K. E. (2017) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1005595 · Cited by 8,653
UniProt: a worldwide hub of protein knowledge
The UniProt Consortium (2019) · Nucleic Acids Research · doi:10.1093/nar/gky1049 · Cited by 7,698
Pfam: The protein families database in 2021
Mistry, J., Chuguransky, S., Williams, L., Qureshi, M., Salazar, G. A., Sonnhammer, E. L. L., Tosatto, S. C. E., Paladin, L., Raj, S., Richardson, L. J., et al. (2021) · Nucleic Acids Research · doi:10.1093/nar/gkaa913 · Cited by 7,452
Accelerated Profile HMM Searches
Eddy, S. R. (2011) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1002195 · Cited by 6,975
Dropout as a bayesian approximation: Representing model uncertainty in deep learning
Gal, Y., & Ghahramani, Z. (2015) · arXiv · doi:10.48550/arxiv.1506.02142 · Cited by 6,471
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
Self-citationMMseqs2 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
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
Research works printed but not linked: 5
- Evans, R. et al. bioRxiv 2021.10.04.463034 (2021).
- (2020), Nucleic Acids Res., 48, D570
- Delmont, T. O. et al. bioRxiv 2020.10.15.341214 (2020).
- Alexander, H. et al. bioRxiv 2021.07.25.453713 (2021).
- Mosalaganti, S. et al. bioRxiv 2021.10.26.465776 (2021).
Cited by
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 5,990
UCSF ChimeraX: Tools for structure building and analysis
Meng, E. C., Goddard, T. D., Pettersen, E. F., Couch, G. S., Pearson, Z. J., Morris, J. H., & Ferrin, T. E. (2023) · Protein Science · doi:10.1002/pro.4792 · Cited by 4,442
Scientific discovery in the age of artificial intelligence.
Wang, H., Fu, T., Du, Y., Gao, W., Huang, K., Liu, Z., Chandak, P., Liu, S., Van Katwyk, P., Deac, A., et al. (2023) · Nature · doi:10.1038/s41586-023-06221-2 · Cited by 1,823
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
Large language models generate functional protein sequences across diverse families
Madani, A., Krause, B., Greene, E. R., Subramanian, S., Mohr, B. P., Holton, J. M., Olmos, J. L., Xiong, C., Sun, Z. Z., Socher, R., et al. (2023) · Nature Biotechnology · doi:10.1038/s41587-022-01618-2 · Cited by 1,078
Metadata
- Authors
- Mirdita, M., Schütze, K., Moriwaki, Y., Heo, L., Ovchinnikov, S., & Steinegger, M.
- Affiliations
- Max Planck Institute for Multidisciplinary Sciences (DE)
- Seoul National University (KR)
- Collaborative Research Institute for Innovative Microbiology, The University of Tokyo (JP)
- Michigan State University (US)
- Harvard University (US)
- Corresponding author
- Mirdita, Milot
- Ovchinnikov, Sergey
- Steinegger, Martin
- ORCID
- Venue
- Nature Methods
- Publisher
- openRxiv
- Published
- 2021-08-15
- Volume
- 19
- Issue
- 6
- Pages
- 679-682
- ISSN
- 1548-7091 (ISSN-L)
- PMC
- PMC9184281
- OpenAlex
- W4281790889
- Cited by
- 9,391
- References
- 40
- Citations by year
- 2026: 1465 · 2025: 2293 · 2024: 2715 · 2023: 2017 · 2022: 817 · 2021: 75
- FWCI
- 644.88
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Nature Methods (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)
- GoeScholar The Publication Server of the Georg-August-Universität Göttingen (Georg-August-Universität Göttingen) (submittedVersion, cc-by)
- Seoul National University Open Repository (Seoul National University) (submittedVersion)
- Funders
- National Science Foundation (2032259)
- Japan Agency for Medical Research and Development (JP21am0101107)
- National Research Foundation (2019R1A6A1A10073437)
- Bundesministerium für Bildung und Forschung
- Seoul National University
- Georg-August-Universität Göttingen
- National Research Foundation of Korea (2021-R1C1-C102065, 2019R1A6A1A10073437, 2020M3A9G7103933, 2021R1C1C102065, 2021M3A9I4021220, 2021-M3A9-I4021220, 2020M3‐A9G7‐103933, 2019R1-A6A1-A10073437)
- Max-Planck-Gesellschaft
- National Institutes of Health (DP5OD026389)
- National Institute of Allergy and Infectious Diseases (R21AI156595)
- 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 structure prediction, protein complex prediction, homology search, MMseqs2, AlphaFold2, RoseTTAFold, Google Colaboratory, GPU acceleration, open source software, environmental databases, ColabFold
- MeSH terms
- Computers, Proteins, Software, Databases, Factual, Protein Folding
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 (qwen3.8-omni-flash and 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: ColabFold - Making protein folding accessible to all. https://app.cochat.ai/papers/10.1101/2021.08.15.456425
- The paper
- Mirdita, M., Schütze, K., Moriwaki, Y., Heo, L., Ovchinnikov, S., & Steinegger, M. (2021). ColabFold - Making protein folding accessible to all. Nature Methods. https://doi.org/10.1101/2021.08.15.456425
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.