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
AI snapshot
Written from the paper’s full text.
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
Claims the paper makes
0 of 19 cited sources checked · 0 supported · 0 flagged
- Findingown data
ColabFold matches AlphaFold2 on CASP14 targets in prediction quality.
“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.”
Quote in the paperSubstantial claim - Findingown data
ColabFold outperforms AlphaFold-Colab on CASP14 targets in prediction quality.
“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.”
Quote in the paperSubstantial claim - Findingreasoning
ColabFold's optimized model utilization enables prediction of close to 1,000 structures per day on a server with one GPU.
“ColabFold’s 40−60-fold faster search and optimized model utilization enables prediction of close to 1,000 structures per day on a server with one graphics processing unit.”
Quote in the paperSubstantial claim - Findingown data
ColabFold matches AlphaFold-multimer on the ClusPro dataset in prediction quality.
“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.”
Quote in the paperSubstantial claim - Findingown data
AlphaFold2 (reduced_dbs, v2.1.1), ColabFold default and ColabFold Stop ≥ 85 have an average pLDDT of 90.68, 90.22 and 89.33, respectively, for 50 randomly sampled 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.”
Quote in the paperSubstantial claim
- Findingown data
For homo-oligomeric complexes, providing a separate MSA copy (padding by gap characters to extend to other copies) works significantly better than concatenating from left to right.
“Interestingly, it was found that providing a separate MSA copy (padding by gap characters to extend to other copies) works significantly better than concatenating from left to right.”
Quote in the paperNot a substantial claim - Findingnone given
When templates are enabled, compiling model 1 and using weights from model 2, and compiling model 3 and using weights from models 4 and 5, saves 5 min of compile time.
“When templates are enabled, model 1 is compiled and weights from model 2 are used, model 3 is compiled and weights from models 4 and 5 are used. This saves 5 min of compile time.”
Quote in the paperSubstantial claim - Findingown data
ColabFold achieves the highest accuracy in the prediction of complexes on the ClusPro dataset with the AlphaFold-multimer model.
“ColabFold achieves the highest accuracy in the prediction of complexes on the ClusPro4,12 dataset with the AlphaFold-multimer model, however, some targets performed better using the residue-index mode.”
Quote in the paperNot a substantial claim - Findingown data
For difficult targets and designed proteins without known homologs, additional recycling iterations can result in a high-quality prediction.
“For difficult targets as well as for designed proteins without known homologs, additional recycling iterations can result in a high-quality prediction (Supplementary Fig. 5).”
Quote in the paperSubstantial claim - Findingown data
ColabFoldDB produces more diverse MSAs than BFD/MGnify for Pfam domains with <30 members.
“In Supplementary Fig. 2 we show that ColabFoldDB, in comparison with BFD/MGnify, produces more diverse MSAs for domains in the protein families database (Pfam)19 with <30 members.”
Quote in the paperSubstantial claim - Findingcitation
Combining two sequences with a glycine linker often successfully models protein complexes in AlphaFold2.
“However, we found that by combining two sequences with a glycine linker20 it could often successfully model complexes.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [20] Moriwaki, Y . AlphaFold2 can also predict heterocomplexes. all you have to do is input the two sequences you want to predict and connect them with a long linker. https://twitter.com/Ag_smith/status/1417063635000598528 (2021).
Not checked in the free run.
- Findingreasoning
The expanding search results in a speed-up of ~10-fold because only 29.3 million cluster consensus sequences are searched instead of all 277.5 million UniRef100 sequences.
“This expanding search results in a speed-up of ~10-fold given that only 29.3 million cluster consensus sequences are searched instead of all 277.5 million UniRef100 sequences.”
Quote in the paperSubstantial claim - Findingcitation
ColabFold complexes were successfully used to aid in the determination of the structure of the 120 MDa human nucleopore complex on cryogenic electron microscopy.
“ColabFold complexes were successfully used to aid in the determination of the structure of the 120 MDa human nucleopore complex on cryogenic electron microscopy23.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [23] Mosalaganti, S. et al. bioRxiv 2021.10.26.465776 (2021).
Not checked in the free run.
- Findingnone given
For predictions of high certainty (pLDDT >85), all five models would often produce structures of very similar confidence, for some even without or with less than three recycles.
“We noted that for prediction of high certainty (pLDDT >85), all five models would often produce structures of very similar confidence, for some even without or with less than three recycles.”
Quote in the paperSubstantial claim - Findingnone given
When no templates are provided, compiling once and replacing the weights from the other models using the configuration of model 5 saves 7 min of compile time.
“When no templates are provided, we compile once and, during inference, replace the weights from the other models, using the configuration of model 5. This saves 7 min of compile time.”
Quote in the paperNot a substantial claim - Findingown data
ColabFold is on average fivefold faster for single predictions than AlphaFold2 and AlphaFold-Colab, taking both MSA generation and model inference into account.
“ColabFold is on average fivefold faster for single predictions than AlphaFold2 and AlphaFold-Colab, when taking both MSA generation (Fig. 2b) and model inference into account.”
Quote in the paperSubstantial claim - Findingown data
For some designed proteins without known homologous sequences, additional recycles helped to fold the final protein.
“For some designed proteins without known homologous sequences, this helped to fold the final protein (Supplementary Fig. 5).”
Quote in the paperNot a substantial claim - Findingnone given
Some complex targets performed better using the residue-index mode.
“ColabFold achieves the highest accuracy in the prediction of complexes on the ClusPro4,12 dataset with the AlphaFold-multimer model, however, some targets performed better using the residue-index mode.”
Quote in the paperNot a substantial claim - Findingnone given
The efforts to speed up AlphaFold2's MSA generation yielded large improvements in its run time.
“Our efforts in speeding up AlphaFold2’s MSA generation yielded large improvements in its run time.”
Quote in the paperSubstantial claim - Findingown data
Over all predictions, the pLDDTs for the M. jannaschii proteome from the AlphaFoldDB, ColabFold default and ColabFold Stop ≥ 85 are 89.75, 89.38 and 88.77, respectively.
“Over all predictions, the pLDDTs for the M. jannaschii proteome downloaded from the AlphaFoldDB, ColabFold default and ColabFold Stop ≥ 85 are 89.75, 89.38 and 88.77, respectively.”
Quote in the paperNot a substantial claim - Findingown data
The structures of the M. jannaschii proteome were predicted on a single Nvidia Titan RTX with 24 GB RAM in 46 h using only MSAs (no templates).
“We then predict the structures on a single Nvidia Titan RTX with 24 GB RAM in 46 h using only MSAs (no templates).”
Quote in the paperSubstantial claim - Findingown data
With the colabfold_search wrapper to MMseqs2, searching the M. jannaschii proteome against ColabFoldDB took 113 min on a system with an AMD EPYC 7402P 24-core CPU and 512 GB RAM.
“With the colabfold_search wrapper to MMseqs2 we search against the ColabFoldDB (section 2.3.2) in 113 min on a system with an AMD EPYC 7402P 24-core CPU (no hyperthreading) and 512 GB RAM.”
Quote in the paperNot a substantial claim - Premisecitation
Prediction quality depends on the input MSA.
“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).”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [1] source
Not checked in the free run.
- Premisecitation
An MSA with only a few (~30) sufficiently diverse sequences is often enough to produce high-quality predictions.
“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).”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [1] source
Not checked in the free run.
- Premisenone given
RoseTTAFold and later AlphaFold were shown to generalize to protein complexes.
“In addition to predictions for single chains, RoseTTAFold and, later, AlphaFold, were also shown to generalize to protein complexes.”
Quote in the paperNot a substantial claim - Premisereasoning
To leverage these methods, researchers require powerful computing capabilities.
“To leverage the power of these methods, researchers require powerful computing capabilities.”
Quote in the paperSubstantial claim- Rests on an assumption: Searches against large environmental databases can take up to hours for a single protein and require more than 2 TB of storage space alone.
- Rests on an assumption: Executing deep neural networks requires GPUs with large GPU RAM even for ~1,000 residue proteins.
- Interpretationnone given
ColabFold combines MMseqs2 homology search with AlphaFold2 or RoseTTAFold to accelerate prediction of protein structures and complexes.
“ColabFold offers accelerated prediction of protein structures and complexes by combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold.”
Quote in the paperSubstantial claim - Interpretationreasoning
The expanding search has the advantage of being more sensitive because cluster consensus sequences are used.
“Additionally, it has the advantage of being more sensitive given that the cluster consensus sequences are used.”
Quote in the paperNot a substantial claim - Premisecitation
A paired MSA helps AlphaFold2 to predict complexes more accurately only if orthologous genes are paired with each other.
“A paired MSA helps AlphaFold2 to predict complexes more accurately only if orthologous genes are paired with each other.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Bryant et al.22 Bryant, P. et al. bioRxiv 2021.09.15.460468 (2021). source
Not checked in the free run.
- Interpretationreasoning
The environmental search database presented an opportunity to improve structure predictions of non-bacterial sequences.
“The environmental search database presented an opportunity to improve structure predictions of non-bacterial sequences given that, for example, eukaryotic protein diversity is not well represented in the databases BFD and MGnify.”
Quote in the paperSubstantial claim- Rests on an assumption: Eukaryotic protein diversity is not well represented in the databases BFD and MGnify.
- Rests on an assumption: Limitations in assembly and gene calling due to complex intron and exon structures result in underrepresentation in reference databases.
- Interpretationreasoning
By offsetting the residue index between two proteins to be greater than 32, AlphaFold2 treats them as separate polypeptide chains.
“By offsetting the residue index between two proteins to be > 32, AlphaFold2 treats them as separate polypeptide chains.”
Quote in the paperSubstantial claim - Interpretationown data
To keep all required sequences and data structures in memory, the size of BFD and MGnify needed to be reduced because both together would have required ~517 GB RAM for headers and sequences alone.
“To keep all required sequences and data structures in memory we needed to reduce the size of the environmental databases BFD and MGnify, given that both databases together would have required ~517 GB RAM for headers and sequences alone.”
Quote in the paperSubstantial claim- Rests on an assumption: AlphaFold2 requires more than 2 TB of storage space for its databases, which is a significant hurdle for many researchers.
- Rests on an assumption: BFD is a clustered protein database consisting of ~2.2 billion proteins organized in 64 million clusters.
- Rests on an assumption: MGnify (2019_05) contains ~300 million environmental proteins.
- Premisecitation
Mirdita et al. improved MMseqs2 to support fast single-against-many searches.
“Mirdita et al.11 improved it to also support fast single-against-many searches.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [11] source
Not checked in the free run.
- Interpretationown data
The inter-PAE, the predicted TM-score or interface TM-score (both derived from PAE) can be used to rank and assess the confidence of a predicted protein-protein interaction.
“As illustrated in Supplementary Fig. 4c, the inter-PAE, the predicted TM-score or interface TM-score (both derived from PAE) can be used to rank and assess the confidence of the predicted protein–protein interaction.”
Quote in the paperSubstantial claim - Interpretationown data
ColabFold does not sacrifice accuracy for its much faster run times, as shown across multiple datasets.
“We show with multiple datasets that ColabFold does not sacrifice accuracy for its much faster run times.”
Quote in the paperSubstantial claim- Rests on an assumption: The benchmark datasets used (CASP14 targets, ClusPro targets and the M. jannaschii proteome) are representative of the protein structure prediction tasks ColabFold is intended for.
- Premisecitation
RoseTTAFold is able to model complexes despite being trained only on single chains.
“Baek et al.3 show that RoseTTAFold is able to model complexes despite being trained only on single chains.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Baek et al.3 source
Not checked in the free run.
- Premisenone given
AlphaFold2 uses relative positional encoding with a cap, meaning that any pair of residues separated by 32 or more are given the same relative positional encoding.
“AlphaFold2 uses relative positional encoding with a cap at , meaning that any pair of residues separated by 32 or more are given the same relative positional encoding.”
Quote in the paperNot a substantial claim
- Classified as probably not a substantial claim. 18
- Borderline: one of the tests for a substantial claim is unclear. 1
- The quote only partly states this claim. 2
- May combine two claims. 17
Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).
Findingnone given§ MainThe prediction of target T1084 can be improved from a TM-score of 0.457 to 0.872 by ColabFold if MMseqs2's compositional filter is disabled.
“The prediction of target T1084 can be improved from a TM-score of 0.457 to 0.872 by ColabFold if MMseqs2’s compositional filter is disabled (Supplementary Fig. 3).”
Quote in the paperNot a substantial claimBorderline on S4 (based): unclear from the text.
Findingnone given§ MainThe average pLDDTs of AlphaFold2 and ColabFold Stop ≥85 were 89.75 and 88.78 in a subsampled set of 50 proteins.
“The average pLDDTs of AlphaFold2 and ColabFold Stop ≥ 85 were 89.75 and 88.78 in a subsampled set of 50 proteins.”
Quote in the paperNot a substantial claimThe quote only partly states this claim.
FindingcitationColabFold's homology search is 40-60-fold faster than AlphaFold2's.
“ColabFold’s 40−60-fold faster search and optimized model utilization enables prediction of close to 1,000 structures per day on a server with one graphics processing unit.”
Quote in the paperSubstantial claim0 of 2 sources checkedMay combine two claims.
Findingown dataEarly stopping results in a speed-up of 3.7-fold compared with the default and 4.8-fold compared with always recompiling.
“Early stopping results in a speed-up of 3.7-fold compared with the default and 4.8-fold compared with always recompiling.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataFiltering each cluster to keep only the 10 most diverse sequences reduced the total number of sequences from 2.5 billion to 513 million, requiring only 84 GB RAM for headers and sequences.
“To reduce the size of the database we filtered each cluster, keeping only the 10 most diverse sequences using mmseqs filterresult --diff 10. This reduced the total number of sequences from 2.5 billion to 513 million, thus requiring only 84 GB RAM for headers and sequences.”
Quote in the paperSubstantial claim- Rests on an assumption: BFD is a clustered protein database consisting of ~2.2 billion proteins organized in 64 million clusters.
- Rests on an assumption: MGnify (2019_05) contains ~300 million environmental proteins.
May combine two claims.
Findingown dataOver all CASP14 targets, TM-scores are 0.887 for ColabFold-AlphaFold2-BFD/MGnify, 0.886 for ColabFold-AlphaFold2-ColabFoldDB, 0.888 for AlphaFold2, and 0.754 for ColabFold-RoseTTAFold-BFD/MGnify.
“Over all CASP14 targets (excluding AlphaFold-Colab because it cannot be used as a standalone) the TM-scores are 0.887, 0.886, 0.888 and 0.754 for the respective methods.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataThe Methanocaldococcus jannaschii proteome of 1,762 proteins shorter than 1,000 amino acids can be predicted in 48 h with early stopping at pLDDT ≥85 on one Nvidia Titan RTX, while sacrificing little or no prediction accuracy.
“All together, we show that the Methanocaldococcus jannaschii proteome of 1,762 proteins shorter than 1,000 amino acids can be predicted in 48 h with early stopping at a pLDDT (predicted local distance difference test; a per-residue confidence metric) of ≥85 on one Nvidia Titan RTX (Fig. 2d), while sacrificing little or no prediction accuracy (Methods 2.7.4).”
Quote in the paperSubstantial claimMay combine two claims.
FindingreasoningSorting input queries by sequence length, processing them in ascending order, and padding input features by 10% avoids recompilation for sequences within the query length plus a 10% margin, resulting in a large speed-up for short proteins.
“We sort the input queries by sequence length and process them in ascending order. We pad the input features by 10% (by default). All sequences that lie within the query length and an additional 10% margin are not required to be recompiled, resulting in a large speed-up for short proteins.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
FindingreasoningChanging the random seed can result in different cluster centers and thus different structure predictions.
“Changing the random seed can result in different cluster centers and thus different structure predictions.”
Quote in the paperNot a substantial claimMay combine two claims.
FindingreasoningIn AlphaFold2, residue index manipulation allows modeling of complexes, and the paired alignment is often not needed.
“In AlphaFold2 we find the same to be true, although surprisingly the paired alignment is often not needed (Fig. 2c).”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataRerunning the CASP14 benchmark with a recycle count of 12 improved targets with little MSA information, resulting in an increased average TM-score of 0.898.
“Rerunning the CASP14 benchmark with a recycle count of 12 resulted in an improvement of targets with little MSA information, resulting in an increased average TM-score of 0.898 (Supplementary Fig. 6).”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataAdditional recycles can often improve a model at the cost of a longer run time.
“We exposed the recycle count as a customizable parameter given that additional recycles can often improve a model (Supplementary Fig. 6) at the cost of a longer run time.”
Quote in the paperSubstantial claimMay combine two claims. The quote only partly states this claim.
Findingown dataThe final ColabFoldDB database consists of 209,335,865 representative sequences and 738,695,580 members.
“The final database consists of 209,335,865 million representative sequences and 738,695,580 members (see the Data Availability section for the input files).”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataMean TM-scores for free-modeling CASP14 targets are 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.
“The mean TM-scores for the free-modeling targets are 0.826, 0.818, 0.79, 0.744 and 0.62 for ColabFold-AlphaFold2-BFD/MGnify, ColabFold-AlphaFold2-ColabFoldDB, AlphaFold2, AlphaFold-Colab and ColabFold-RoseTTAFold-BFD/MGnify, respectively.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataMMseqs2 had a maximum resident set size of 308 GB during the ColabFoldDB search.
“MMseqs2 had a maximum resident set size of 308 GB during the search.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Premisecitation§ MainAlphaFold2 predicted 3D atomic coordinates of folded protein structures at a median GDT_TS of 92.4% in CASP14.
“AlphaFold2 (ref. 1) was able to predict the 3D atomic coordinates of folded protein structures at a median global distance test total score (GDT_TS) of 92.4% in the latest round of the protein folding competition by the international community, CASP14 (Critical Assessment of protein Structure Prediction, round 14) (ref. 2).”
Quote in the paperNot a substantial claim0 of 2 sources checked- cites ref. 1
Not checked in the free run.
- cites ref. 2
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Premisecitation§ MainEvans et al. released AlphaFold-multimer, a refined version of AlphaFold2 for protein complex prediction.
“Evans et al.4 have since released AlphaFold-multimer, a refined version of AlphaFold2 for the prediction of protein complexes.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Evans et al.4 Evans, R. et al. bioRxiv 2021.10.04.463034 (2021).
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Premisecitation§ MainFor high-quality predictions, sequences should be provided in paired form to AlphaFold2.
“For high-quality predictions it was shown that sequences should be provided in paired form to AlphaFold2 (ref. 22).”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites ref. 22
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Premisecitation§ MainMany ideas of AlphaFold2 were independently reproduced and implemented in RoseTTAFold.
“Many ideas of AlphaFold2 were independently reproduced and implemented in RoseTTAFold (ref. 3).”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites ref. 3
Not checked in the free run.
Probably not a substantial claim: may not pass S4 (based).
Interpretationnone given§ AbstractCoupled with Google Colaboratory, ColabFold becomes a free and accessible platform for protein folding.
“Coupled with Google Colaboratory, ColabFold becomes a free and accessible platform for protein folding.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
PremisecitationEnvironmental databases contain billions of proteins extracted from metagenomic and transcriptomic experiments, which often complement databases dominated by isolated genomes.
“These environmental databases contain billions of proteins extracted from metagenomic and transcriptomic experiments, which often complement databases dominated by isolated genomes.”
Quote in the paperNot a substantial claim0 of 2 sources checked- cites [1] source
Not checked in the free run.
- cites [6] (2020), Nucleic Acids Res.
Not checked in the free run.
May combine two claims.
PremisecitationAlphaFold-Colab results in less accurate predictions while still requiring long search times.
“This results in less accurate predictions while still requiring long search times.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [9] (2021), Nature source
Not checked in the free run.
May combine two claims.
PremisecitationTunyasuvunakool et al. developed AlphaFold-Colab, an AlphaFold2 Jupyter Notebook for Google Colaboratory, with input MSA built by HMMer against UniRef90 and an eightfold-reduced environmental database.
“Concurrently, Tunyasuvunakool et al.9 developed an AlphaFold2 Jupyter Notebook for Google Colaboratory (referred to as AlphaFold-Colab), for which the input MSA is built by searching with HMMer against the UniProt Reference Clusters (UniRef90) and an eightfold-reduced environmental database.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [9] (2021), Nature source
Not checked in the free run.
May combine two claims.
InterpretationreasoningTwo highly accurate open-source prediction methods for single chains and one for protein complexes are now publicly available.
“Thus, two highly accurate open-source prediction methods for single chains and one for protein complexes are now publicly available.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Premisenone givenMMseqs2 was initially built to perform fast many-against-many sequence searches.
“MMseqs2 was initially built to perform fast many-against-many sequence searches.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationIncreasing the model's internal parameter residue-index can also be done in AlphaFold2.
“Shortly afterwards, Baek21 found that increasing the model's internal parameter, residue-index (the method that was used in RoseTTAFold), could also be done in AlphaFold2.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [21] Baek, M . Adding a big enough number for “residue_index” feature is enough to model hetero-complex using AlphaFold (green&cyan: crystal structure / magenta: predicted model w/ residue_index modification). https://twitter.com/minkbaek/status/1417538291709071362 (2021).
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Interpretationnone givenThe inter-chain predicted alignment error (inter-PAE) provided by AlphaFold2 helps to rank complexes.
“The inter-chain predicted alignment error (inter-PAE) provided by AlphaFold2 helps to rank the complexes.”
Quote in the paperNot a substantial claimMay combine two claims.
InterpretationreasoningColabFold builds beyond AlphaFold2 by improving sequence search, providing tools for homo- and heteromer complexes, exposing advanced functionality, expanding environmental databases, and enabling large-scale batch prediction at ~90-fold speed-up.
“ColabFold builds beyond the initial offerings of Alphafold2 by improving its sequence search, providing tools for modeling homo- and heteromer complexes, exposing advanced functionality, expanding the environmental databases and enabling large-scale batch prediction of protein structures, at an approximately 90-fold speed-up over AlphaFold2.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductnone givenMSA generation was repeated five times and the MSA generation times were averaged.
“MSA generation was repeated five times and the MSA generation times were averaged.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Premisenone givenAlphaFold2, depending on model configuration, subsamples the MSA to a maximum of 512 cluster centers and 1,024 extra sequences.
“Alphafold2, depending on model configuration, subsamples the MSA to a maximum of 512 cluster centers and 1,024 extra sequences.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Interpretationnone givenThe 2D renderer's advantage over a 3D renderer is that its images are very lightweight, can be used in animations and saved as vector graphics for lossless inclusion in documents.
“The advantage over a 3D renderer is that the images are very lightweight, can be used in animations and saved as vector graphics for lossless inclusion in documents.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductnone givenFor a fair comparison, AlphaFold2 was modified to allow HMMer and HHblits to access one CPU core.
“For a fair comparison, AlphaFold2 was modified to allow HMMer and HHblits to access one CPU core.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Premisenone givenAlphaFold2 improves the predicted protein structure by recycling (by default) three times, meaning that the prediction is fed multiple times through the model.
“AlphaFold2 improves the predicted protein structure by recycling (by default) three times, meaning that the prediction is fed multiple times through the model.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningEnabling is_training (the stochastic dropout part of the model) can be used to sample an ensemble of models for the uncertain parts of the structure prediction.
“Enabling the latter can be used to sample an ensemble of models for the uncertain parts of the structure prediction.”
Quote in the paperNot a substantial claim
The paper relies on this without support.
Searches against large environmental databases can take up to hours for a single protein and require more than 2 TB of storage space alone.
“Due to their large size, searches can take up to hours for a single protein while requiring more than 2 TB of storage space alone.”
The paper relies on this without support.
Executing deep neural networks requires GPUs with large GPU RAM even for ~1,000 residue proteins.
“Second, to execute the deep neural networks, graphics processing units (GPUs) with a large amount of GPU RAM (random access memory) are required even for relatively common protein sizes of ~1,000 residues.”
The paper relies on this without support.
For these proteins, MSA generation dominates the overall run time.
“For these, however, the MSA generation dominates the overall run time.”
The paper relies on this without support.
Eukaryotic protein diversity is not well represented in the databases BFD and MGnify.
“The environmental search database presented an opportunity to improve structure predictions of non-bacterial sequences given that, for example, eukaryotic protein diversity is not well represented in the databases BFD and MGnify.”
The paper relies on this without support.
Limitations in assembly and gene calling due to complex intron and exon structures result in underrepresentation in reference databases.
“Limitations in assembly and gene calling due to complex intron and exon structures result in underrepresentation in reference databases.”
The paper relies on this without support.
AlphaFold2 was initially released without the capability to model protein complexes.
“AlphaFold2 was initially released without the capability to model protein complexes.”
The paper relies on this without support.
Generating MSAs for AlphaFold2 and RoseTTAFold is a time-consuming task.
“Generating MSAs for AlphaFold2 and RoseTTAFold is a time-consuming task.”
The paper relies on this without support.
This type of search requires the database to be indexed and stored in memory.
“This type of search requires the database to be indexed and stored in memory.”
The paper relies on this without support.
AlphaFold2 requires more than 2 TB of storage space for its databases, which is a significant hurdle for many researchers.
“AlphaFold2 requires more than 2 TB of storage space for its databases, which is a significant hurdle for many researchers.”
The paper relies on this without support.
BFD is a clustered protein database consisting of ~2.2 billion proteins organized in 64 million clusters.
“BFD is a clustered protein database consisting of ~2.2 billion proteins organized in 64 million clusters.”
The paper relies on this without support.
MGnify (2019_05) contains ~300 million environmental proteins.
“MGnify (2019_05) contains ~300 million environmental proteins.”
The paper relies on this without support.
The benchmark datasets used (CASP14 targets, ClusPro targets and the M. jannaschii proteome) are representative of the protein structure prediction tasks ColabFold is intended for.
“We show with multiple datasets that ColabFold does not sacrifice accuracy for its much faster run times.”
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?
AlphaFold-Multimer predicts the SALL4 QID forms an alpha-helix with high confidence and assembles as a four-helical bundle.
“The QID is predicted to form an alpha-helix with high confidence and all five models suggest higher order assembly as a four-helical bundle with almost perfect overlap (Figure 1F and Figure S2C,D).”
From Tetramerisation governs SALL transcription factor function in development and disease (2025) · cites this paper as [37]
The Methods section, “Modeling protein complexes with ColabFold” (page not available), describes using AlphaFold-Multimer to predict protein complexes. Searches for SALL4, QID, an alpha-helix, and a four-helical bundle found no passage reporting the specific prediction or the agreement among five models. (LLM, conf=0.76)
Experimental — accuracy not yet measuredColabFold and OpenFold let users run a modified AF2 protocol on open-access servers within minutes.
“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]
For targets without known homologs, increasing recycling iterations can improve prediction quality.
“For targets without known homologs, such as computationally designed proteins, increasing the number of recycling iterations can improve the quality of the prediction8”
From The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins (2024) · cites this paper as [8]
Enabling dropout layers in the neural network, normally used only during training, is another approach to increase conformational sampling.
“Another approach is to enable dropout layers in the neural network, which are usually commonly used only during neural network training8,106,107.”
From The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins (2024) · cites this paper as [8]
Contact predictions can be used to simultaneously fold and dock proteins, and AF2 can predict the structure of complexes.
“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]
AF2 run using ColabFold predicts the thermodynamically unfavoured FS state for KaiBTE.
“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]
Modifying the residue_index feature used by AF2 eliminates the need for a covalent linker that likely creates artifacts.
“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]
A homologous protein 64% identical to UniProt A0A1C5UEQ5 from a cultured Lachnospiraceae bacterium was predicted with ColabFold and confirmed to have a similar DNA-binding domain structure (TM score 0.97 and 0.56 in relation to UniProt A0A1C5UEQ5 and human AIM, respectively).
“We predicted the structure of one homologous protein that is 64% identical to UniProt A0A1C5UEQ5 (Extended Data Fig. 8)—which originates from a cultured Lachnospiraceae bacterium that is part of the Culturable Genome Reference30 of the human gut—using ColabFold31 and confirmed that it has a similar structure DNA-binding domain structure (TM score of 0.97 and 0.56 in relation to UniProt A0A1C5UEQ5”
From Clustering predicted structures at the scale of the known protein universe (2023) · cites this paper as [31]
AlphaFold and RoseTTAfold have demonstrated unprecedented accuracy in predicting structures of complexes.
“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]
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.
Reported pLDDT averages for the 50-protein sample disagree
MinorThe main text reports averages of 89.75 for AlphaFold2 and 88.78 for ColabFold Stop ≥ 85 in a sample of 50 proteins. The Methods section reports 90.68 and 89.33, respectively, for 50 randomly sampled proteins.
“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.”
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
Self-citationMatplotlib: 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
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 8,728 · see citing works
UniProt: a worldwide hub of protein knowledge
The UniProt Consortium (2019) · Nucleic Acids Research · doi:10.1093/nar/gky1049 · Cited by 7,698 · see citing works
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 · see citing works
Accelerated Profile HMM Searches
Eddy, S. R. (2011) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1002195 · Cited by 6,975 · see citing works
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 · see citing works
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 5,542 · see citing works
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 · 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
TM-align: a protein structure alignment algorithm based on the TM-score
Zhang, Y. (2005) · Nucleic Acids Research · doi:10.1093/nar/gki524 · Cited by 3,774 · see citing works
The ClusPro web server for protein–protein docking
Kozakov, D., Hall, D. R., Xia, B., Porter, K. A., Padhorny, D., Yueh, C., Beglov, D., & Vajda, S. (2017) · Nature Protocols · doi:10.1038/nprot.2016.169 · Cited by 3,297 · see citing works
Jupyter Notebooks – a publishing format for reproducible computational workflows
Kluyver, T. A., Ragan-Kelley, B., Fernando, P., Granger, B., Bussonnier, M., Frederic, J., Kelley, K., Hamrick, J. B., Grout, J., Corlay, S., et al. (2016) · IOS Press eBooks · doi:10.3233/978-1-61499-649-1-87 · Cited by 3,191 · 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 1× · see citing works
“Concurrently, Tunyasuvunakool et al. 9 developed an AlphaFold2 Jupyter Notebook for Google Colaboratory (referred to as AlphaFold-Colab), for which the input MSA is built by searching with HMMer against the UniProt Reference Clusters (UniRef90) and an eightfold-reduced environmental database.”
MethodOpenMM 7: Rapid development of high performance algorithms for molecular dynamics
Eastman, P., Swails, J., Chodera, J. D., McGibbon, R. T., Zhao, Y., Beauchamp, K. A., Wang, L.-P., Simmonett, A. C., Harrigan, M. P., Stern, C. D., et al. (2017) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1005659 · Cited by 2,748 · see citing works
UniRef clusters: a comprehensive and scalable alternative for improving sequence similarity searches
Suzek, B. E., Wang, Y., Huang, H., McGarvey, P. B., Wu, C. H., & the UniProt Consortium (2015) · Bioinformatics · doi:10.1093/bioinformatics/btu739 · Cited by 1,928 · see citing works
Federhen, S. (2012) · Nucleic Acids Research · doi:10.1093/nar/gkr1178 · Cited by 1,436 · see citing works
Improved prediction of protein side‐chain conformations with SCWRL4
Krivov, G. G., Shapovalov, M. V., & Dunbrack, R. L. (2009) · Proteins Structure Function and Bioinformatics · doi:10.1002/prot.22488 · Cited by 1,312 · see citing works
HH-suite3 for fast remote homology detection and deep protein annotation
Steinegger, M., Meier, M., Mirdita, M., Vöhringer, H., Haunsberger, S. J., & Söding, J. (2019) · BMC Bioinformatics · doi:10.1186/s12859-019-3019-7 · Cited by 1,145 · see citing works
Self-citationImproved 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
Clustering huge protein sequence sets in linear time
Steinegger, M., & Söding, J. (2018) · Nature Communications · doi:10.1038/s41467-018-04964-5 · Cited by 875 · see citing works
Self-citationPyRosetta: a script-based interface for implementing molecular modeling algorithms using Rosetta
Chaudhury, S., Lyskov, S., & Gray, J. J. (2010) · Bioinform. · doi:10.1093/bioinformatics/btq007 · Cited by 772 · see citing works
Uniclust databases of clustered and deeply annotated protein sequences and alignments
Mirdita, M., von den Driesch, L., Galiez, C., Martin, M. J., Söding, J., & Steinegger, M. (2016) · Nucleic Acids Research · doi:10.1093/nar/gkw1081 · Cited by 710 · see citing works
Self-citationMMseqs2 desktop and local web server app for fast, interactive sequence searches
Mirdita, M., Steinegger, M., & Söding, J. (2019) · Bioinformatics · doi:10.1093/bioinformatics/bty1057 · Cited by 618 · see citing works
Self-citationMassive expansion of human gut bacteriophage diversity
Camarillo-Guerrero, L. F., Almeida, A., Rangel-Pineros, G., Finn, R. D., & Lawley, T. D. (2021) · Cell · doi:10.1016/j.cell.2021.01.029 · Cited by 594 · see citing works
Metagenomic compendium of 189,680 DNA viruses from the human gut microbiome
Nayfach, S., Páez-Espino, D., Call, L., Low, S. J., Sberro, H., Ivanova, N. N., Proal, A. D., Fischbach, M. A., Bhatt, A. S., Hugenholtz, P., et al. (2021) · Nature Microbiology · doi:10.1038/s41564-021-00928-6 · Cited by 580 · see citing works
MGnify: the microbiome analysis resource in 2020
Mitchell, A. L., Almeida, A., Beracochea, M., Boland, M., Burgin, J., Cochrane, G., Crusoe, M. R., Kale, V., Potter, S. C., Richardson, L. J., et al. (2019) · Nucleic Acids Research · doi:10.1093/nar/gkz1035 · Cited by 562 · see citing works
Critical assessment of methods of protein structure prediction (CASP)—Round <scp>XIV</scp>
Kryshtafovych, A., Schwede, T., Topf, M., Fidelis, K., & Moult, J. (2021) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.26237 · Cited by 503 · see citing works
DockQ: A Quality Measure for Protein-Protein Docking Models
Basu, S., & Wallner, B. (2016) · PLOS ONE · doi:10.1371/journal.pone.0161879 · Cited by 460 · see citing works
3Dmol.js: molecular visualization with WebGL
Rego, N. B., & Koes, D. R. (2014) · Bioinformatics · doi:10.1093/bioinformatics/btu829 · Cited by 367 · see citing works
Artificial intelligence reveals nuclear pore complexity
Mosalaganti, S., Obarska-Kosinska, A., Siggel, M., Turonova, B., Zimmerli, C. E., Buczak, K., Schmidt, F. H., Margiotta, E., Mackmull, M.-T., Hagen, W., et al. (2021) · doi:10.1101/2021.10.26.465776 · Cited by 355 · see citing works
AI-based structure prediction empowers integrative structural analysis of human nuclear pores
Mosalaganti, S., Obarska-Kosinska, A., Siggel, M., Taniguchi, R., Turoňová, B., Zimmerli, C. E., Buczak, K., Schmidt, F. H., Margiotta, E., Mackmull, M.-T., et al. (2022) · Science · doi:10.1126/science.abm9506 · Cited by 355 · see citing works
Levy Karin, E., Mirdita, M., & Söding, J. (2020) · Microbiome · doi:10.1186/s40168-020-00808-x · Cited by 299 · see citing works
Self-citationDelmont, T. O., Gaia, M., Hinsinger, D. D., Frémont, P., Vanni, C., Fernandez-Guerra, A., Eren, A. M., Kourlaiev, A., d'Agata, L., Clayssen, Q., et al. (2022) · Cell Genomics · doi:10.1016/j.xgen.2022.100123 · Cited by 173 · see citing works
Alexander, H., Hu, S. K., Krinos, A. I., Pachiadaki, M., Tully, B. J., Neely, C. J., & Reiter, T. (2021) · doi:10.1101/2021.07.25.453713 · Cited by 67 · see citing works
Alexander, H., Hu, S. K., Krinos, A. I., Pachiadaki, M., Tully, B., Neely, C. J., & Reiter, T. E. (2023) · mBio · doi:10.1128/mbio.01676-23 · Cited by 66 · see citing works
Cluster Analysis of Stations Based on Weight SimRank in Sharing Bicycle
Guan, E. A. B. (2021) · CONVERTER · doi:10.17762/converter.152 · Cited by 2 · see citing works
Flagged first, then by citations
3 of the 40 references couldn’t be matched to a known paper.
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
Sorted by citations
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
UCSF ChimeraX : Tools for structure building and analysis
Meng, E. C., Goddard, T. D., Pettersen, E. F., Couch, G. S., Pearson, Z., Morris, J. H., & Ferrin, T. E. (2023) · Protein Science · doi:10.1002/pro.4792 · Cited by 4,575 · see citing works
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 · see citing works
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 · see citing works
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 · see citing works
The Galaxy platform for accessible, reproducible, and collaborative data analyses: 2024 update
Community, T. G., Abueg, L. A. L., Afgan, E., Allart, O., Awan, A., Bacon, W., Baker, D., Bassetti, M. E., Batut, B., Bernt, M., et al. (2024) · Nucleic Acids Research · doi:10.1093/nar/gkae410 · Cited by 983 · see citing works
Boltz-2: Towards Accurate and Efficient Binding Affinity PredictionPreprint
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 848 · see citing works
ProtGPT2 is a deep unsupervised language model for protein design
Ferruz, N., Schmidt, S., & Höcker, B. (2022) · Nature Communications · doi:10.1038/s41467-022-32007-7 · Cited by 795 · see citing works
Automated model building and protein identification in cryo-EM maps
Jamali, K., Käll, L., Zhang, R., Brown, A., Kimanius, D., & Scheres, S. H. W. (2024) · Nature · doi:10.1038/s41586-024-07215-4 · Cited by 686 · see citing works
Automated model building and protein identification in cryo-EM maps
Jamali, K., Käll, L., Zhang, R., Brown, A., Kimanius, D., & Scheres, S. H. W. (2023) · doi:10.1101/2023.05.16.541002 · Cited by 686 · see citing works
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 · see citing works
Self-citationThe HADDOCK2.4 web server for integrative modeling of biomolecular complexes
Honorato, R. V., Trellet, M. E., Jiménez‐García, B., Schaarschmidt, J., Giulini, M., Reys, V., Koukos, P. I., Rodrigues, J., Karaca, E., van Zundert, G. C. P., et al. (2024) · Nature Protocols · doi:10.1038/s41596-024-01011-0 · Cited by 653 · see citing works
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 · 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 · see citing works
Self-citationBoltz-1 Democratizing Biomolecular Interaction ModelingPreprint
Wohlwend, J., Corso, G., Passaro, S., Getz, N., Reveiz, M., Leidal, K., Swiderski, W., Atkinson, L., Portnoi, T., Chinn, I., et al. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.11.19.624167 · Cited by 535 · see citing works
Sampling alternative conformational states of transporters and receptors with AlphaFold2
del Alamo, D., Sala, D., Mchaourab, H. S., & Meiler, J. (2022) · eLife · doi:10.7554/elife.75751 · Cited by 511 · see citing works
“0.1 and a modified version of ColabFold ( Mirdita et al., 2021 ) that is available at https://github.com/delalamo/af2_conformations, ( Del Alamo, 2021b copy archived at swh:1:rev:d60db86886186e80622deaa91045caccaf4103d3).”
BackgroundInfluentialChai-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 · see citing works
Ahdritz, 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 · see citing works
“By analyzing intermediate structures produced during training, we also gain insights into the hierarchical manner in which OpenFold learns to fold.”
Improving de novo protein binder design with deep learning
Bennett, N. R., Coventry, B., Goreshnik, I., Huang, B., Allen, A., Vafeados, D., Peng, Y. P., Dauparas, J., Baek, M., Stewart, L., et al. (2023) · Nature Communications · doi:10.1038/s41467-023-38328-5 · Cited by 402 · see citing works
High-resolution <i>de novo</i> structure prediction from primary sequence
Wu, R., Ding, F., Wang, R., Shen, R., Zhang, X., Luo, S., Su, C., Wu, Z., Xie, Q., Berger, B., et al. (2022) · doi:10.1101/2022.07.21.500999 · Cited by 392 · see citing works
Gaurav, A., Bakht, P., Saini, M., Pandey, S., & Pathania, R. (2023) · Microbiology · doi:10.1099/mic.0.001333 · Cited by 392 · see citing works
Phase separation of FSP1 promotes ferroptosis
Nakamura, T., Hipp, C., Mourão, A., Borggräfe, J., Aldrovandi, M., Henkelmann, B., Wanninger, J., Mishima, E., Lytton, E., Emler, D., et al. (2023) · Nature · doi:10.1038/s41586-023-06255-6 · Cited by 365 · see citing works
Clustering predicted structures at the scale of the known protein universe
Barrio-Hernandez, I., Yeo, J., Jänes, J., Mirdita, M., Gilchrist, C. L. M., Wein, T., Varadi, M., Velankar, S., Beltrao, P., & Steinegger, M. (2023) · Nature · doi:10.1038/s41586-023-06510-w · Cited by 362 · see citing works
Self-citationAI-based structure prediction empowers integrative structural analysis of human nuclear pores
Mosalaganti, S., Obarska-Kosinska, A., Siggel, M., Taniguchi, R., Turoňová, B., Zimmerli, C. E., Buczak, K., Schmidt, F. H., Margiotta, E., Mackmull, M.-T., et al. (2022) · Science · doi:10.1126/science.abm9506 · Cited by 355 · see citing works
Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies
Ruffolo, J. A., Chu, L., Mahajan, S. P., & Gray, J. J. (2023) · Nature Communications · doi:10.1038/s41467-023-38063-x · Cited by 339 · see citing works
Scalable emulation of protein equilibrium ensembles with generative deep learning
Lewis, S., Hempel, T., Jiménez-Luna, J., Gastegger, M., Xie, Y., Foong, A. Y. K., Satorras, V. G., Abdin, O., Veeling, B. S., Zaporozhets, I., et al. (2025) · Science · doi:10.1126/science.adv9817 · Cited by 335 · see citing works
Benchmarking <scp>AlphaFold</scp> for protein complex modeling reveals accuracy determinants
Yin, R., Feng, B. Y., Varshney, A., & Pierce, B. G. (2022) · Protein Science · doi:10.1002/pro.4379 · Cited by 335 · see citing works
Benchmarking AlphaFold for protein complex modeling reveals accuracy determinants
Yin, R., Feng, B. Y., Varshney, A., & Pierce, B. G. (2021) · doi:10.1101/2021.10.23.465575 · Cited by 335 · see citing works
ImmuneBuilder: Deep-Learning models for predicting the structures of immune proteins
Abanades, B., Wong, W. K., Boyles, F., Georges, G. J., Bujotzek, A., & Deane, C. M. (2023) · Communications Biology · doi:10.1038/s42003-023-04927-7 · Cited by 316 · see citing works
Bilingual language model for protein sequence and structure
Heinzinger, M., Weissenow, K., Sanchez, J. G., Henkel, A., Mirdita, M., Steinegger, M., & Rost, B. (2024) · NAR Genomics and Bioinformatics · doi:10.1093/nargab/lqae150 · Cited by 311 · see citing works
Self-citationBefore and after AlphaFold2: An overview of protein structure prediction
Bertoline, L. M. F., Lima, A. N., Krieger, J. E., & Teixeira, S. K. (2023) · Frontiers in Bioinformatics · doi:10.3389/fbinf.2023.1120370 · Cited by 309 · see citing works
Tubiana, J., Schneidman-Duhovny, D., & Wolfson, H. J. (2022) · Nature Methods · doi:10.1038/s41592-022-01490-7 · Cited by 274 · see citing works
Neugebauer, M. E., Hsu, A., Arbab, M., Krasnow, N. A., McElroy, A. N., Pandey, S., Doman, J. L., Huang, T. P., Raguram, A., Banskota, S., et al. (2022) · Nature Biotechnology · doi:10.1038/s41587-022-01533-6 · Cited by 273 · see citing works
SPEACH_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
Mechanisms and pathology of protein misfolding and aggregation
Louros, N. N., Schymkowitz, J., & Rousseau, F. (2023) · Nature reviews. Molecular cell biology · doi:10.1038/s41580-023-00647-2 · Cited by 221 · see citing works
Li, C., Stražar, M., Mohamed, A. M., Ávila-Pacheco, J., Walker, R. L., Lebar, T., Zhao, S., Lockart, J., Dame, A., Thurimella, K., et al. (2024) · Cell · doi:10.1016/j.cell.2024.03.014 · Cited by 260 · see citing works
AF2Complex predicts direct physical interactions in multimeric proteins with deep learning
Gao, M., Nakajima An, D., Parks, J. M., & Skolnick, J. (2022) · Nature Communications · doi:10.1038/s41467-022-29394-2 · Cited by 252 · see citing works
Systematic identification of conditionally folded intrinsically disordered regions by AlphaFold2
Alderson, T. R., Pritišanac, I., Kolarić, Đ., Moses, A. M., & Forman-Kay, J. D. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2304302120 · Cited by 240 · see citing works
Programmable protein delivery with a bacterial contractile injection system
Kreitz, J., Friedrich, M., Guru, A., Lash, B., Saito, M., Macrae, R. K., & Zhang, F. (2023) · Nature · doi:10.1038/s41586-023-05870-7 · Cited by 236 · see citing works
PADLOC: a web server for the identification of antiviral defence systems in microbial genomes
Payne, L. J., Meaden, S., Mestre, M. R., Palmer, C., Toro, N., Fineran, P. C., & Jackson, S. A. (2022) · Nucleic Acids Research · doi:10.1093/nar/gkac400 · Cited by 227 · see citing works
Targeted protein degradation via intramolecular bivalent glues
Hsia, O., Hinterndorfer, M., Cowan, A. D., Iso, K., Ishida, T., Sundaramoorthy, R., Nakasone, M. A., Imrichova, H., Schätz, C., Rukavina, A., et al. (2024) · Nature · doi:10.1038/s41586-024-07089-6 · Cited by 222 · see citing works
Multi‐state modeling of G‐protein coupled receptors at experimental accuracy
Heo, L., & Feig, M. (2022) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.26382 · Cited by 215 · see citing works
Self-citationFanzor is a eukaryotic programmable RNA-guided endonuclease
Saito, M., Xu, P., Faure, G., Maguire, S., Kannan, S., Altae-Tran, H., Vo, C.-D.-T., Desimone, A., Macrae, R. K., & Zhang, F. (2023) · Nature · doi:10.1038/s41586-023-06356-2 · Cited by 215 · see citing works
Machine Learning-Guided Protein Engineering
Kouba, P., Kohout, P., Haddadi, F., Bushuiev, A., Samusevich, R., Sedlář, J., Damborský, J., Pluskal, T., Šivic, J., & Mazurenko, S. (2023) · ACS Catalysis · doi:10.1021/acscatal.3c02743 · Cited by 214 · see citing works
Vidović, M., & Rikalović, M. (2022) · Cells · doi:10.3390/cells11111732 · Cited by 210 · see citing works
Design of highly functional genome editors by modelling CRISPR–Cas sequences
Ruffolo, J. A., Nayfach, S., Gallagher, J. P., Bhatnagar, A., Beazer, J., Hussain, R., Russ, J., Yip, J., Hill, E., Pačesa, M., et al. (2025) · Nature · doi:10.1038/s41586-025-09298-z · Cited by 212 · see citing works
Gan, Z. Y., Callegari, S., Cobbold, S. A., Cotton, T. R., Mlodzianoski, M. J., Schubert, A. F., Geoghegan, N. D., Rogers, K. L., Leis, A. P., Dewson, G., et al. (2021) · Nature · doi:10.1038/s41586-021-04340-2 · Cited by 209 · see citing works
Pandey, S., Gao, X. D., Krasnow, N. A., McElroy, A. N., Tao, Y., Duby, J. E., Steinbeck, B. J., McCreary, J., Pierce, S. E., Tolar, J., et al. (2024) · Nature Biomedical Engineering · doi:10.1038/s41551-024-01227-1 · Cited by 207 · see citing works
How good are AlphaFold models for docking-based virtual screening?
Scardino, V., Di Filippo, J. I., & Cavasotto, C. N. (2023) · iScience · doi:10.1016/j.isci.2022.105920 · Cited by 207 · see citing works
Protein structure generation via folding diffusion
Wu, K., Yang, K., van den Berg, R., Alamdari, S., Zou, J., Lu, A. X., & Amini, A. P. (2024) · Nature Communications · doi:10.1038/s41467-024-45051-2 · Cited by 204 · see citing works
Evaluation of AlphaFold-Multimer prediction on multi-chain protein complexes
Zhu, W., Shenoy, A., Kundrotas, P., & Elofsson, A. (2023) · Bioinformatics · doi:10.1093/bioinformatics/btad424 · Cited by 202 · see citing works
Meller, A., Ward, M. D., Borowsky, J. H., Kshirsagar, M., Lotthammer, J. M., Oviedo, F., Ferres, J. L., & Bowman, G. R. (2023) · Nature Communications · doi:10.1038/s41467-023-36699-3 · Cited by 201 · see citing works
Ankh ☥: Optimized Protein Language Model Unlocks General-Purpose ModellingPreprint
Elnaggar, A., Essam, H., Salah-Eldin, W., Moustafa, W., Elkerdawy, M., Rochereau, C., & Rost, B. (2023) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2023.01.16.524265 · Cited by 201 · 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., García Melani, A., Novomisky Nechcoff, J., Salas, M. N., et al. (2022) · Bioinformatics · doi:10.1093/bioinformatics/btac202 · Cited by 196 · see citing works
High-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
Structure of cytoplasmic ring of nuclear pore complex by integrative cryo-EM and AlphaFold
Fontana, P., Dong, Y., Pi, X., Tong, A. B., Hecksel, C. W., Wang, L., Fu, T.-M., Bustamante, C., & Wu, H. (2022) · Science · doi:10.1126/science.abm9326 · Cited by 195 · see citing works
Discovery of phage determinants that confer sensitivity to bacterial immune systems
Stokar-Avihail, A., Fedorenko, T., Hör, J., Garb, J., Leavitt, A., Millman, A., Shulman, G., Wojtania, N., Melamed, S., Amitai, G., et al. (2023) · Cell · doi:10.1016/j.cell.2023.02.029 · Cited by 176 · see citing works
Transformer-based deep learning for predicting protein properties in the life sciences
Chandra, A. A., Tünnermann, L., Löfstedt, T., & Gratz, R. (2023) · eLife · doi:10.7554/elife.82819 · Cited by 192 · see citing works
Elfmann, C., & Stülke, J. (2023) · Nucleic Acids Research · doi:10.1093/nar/gkad350 · Cited by 191 · see citing works
Shuoker, B., Pichler, M. J., Jin, C., Sakanaka, H., Wu, H., Gascueña, A. M., Liu, J., Nielsen, T. S., Holgersson, J., Karlsson, E. N., et al. (2023) · Nature Communications · doi:10.1038/s41467-023-37533-6 · Cited by 190 · see citing works
Virological characteristics of the SARS-CoV-2 Omicron BA.2.75 variant
Saito, A., Tamura, T., Zahradník, J., Deguchi, S., Tabata, K., Anraku, Y., Kimura, I., Ito, J., Yamasoba, D., Nasser, H., et al. (2022) · Cell Host & Microbe · doi:10.1016/j.chom.2022.10.003 · Cited by 186 · 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
Modeling conformational states of proteins with AlphaFold
Sala, D., Engelberger, F., Mchaourab, H. S., & Meiler, J. (2023) · Current Opinion in Structural Biology · doi:10.1016/j.sbi.2023.102645 · Cited by 182 · see citing works
State-specific protein–ligand complex structure prediction with a multiscale deep generative model
Qiao, Z., Nie, W., Vahdat, A., Miller, T. F., & Anandkumar, A. (2024) · Nature Machine Intelligence · doi:10.1038/s42256-024-00792-z · Cited by 178 · see citing works
Tumor-reactive antibodies evolve from non-binding and autoreactive precursors
Mazor, R. D., Nathan, N., Gilboa, A., Stoler‐Barak, L., Moss, L., Solomonov, I., Hanuna, A., Divinsky, Y., Shmueli, M. D., Hezroni, H., et al. (2022) · Cell · doi:10.1016/j.cell.2022.02.012 · Cited by 176 · see citing works
Visualization of translation and protein biogenesis at the ER membrane
Gemmer, M., Chaillet, M. L., van Loenhout, J., Arenas, R. C., Vismpas, D., Gröllers‐Mulderij, M., Koh, F., Albanese, P., Scheltema, R. A., Howes, S. C., et al. (2023) · Nature · doi:10.1038/s41586-022-05638-5 · Cited by 176 · see citing works
Chen, B., Cheng, X., Li, P., Geng, Y., Gong, J., Li, S., Bei, Z., Tan, X., Wang, B., Zeng, X., et al. (2025) · Nature Methods · doi:10.1038/s41592-025-02636-z · Cited by 174 · see citing works
AlphaFold2: A Role for Disordered Protein/Region Prediction?
Wilson, C. J., Choy, W.-Y., & Karttunen, M. (2022) · International Journal of Molecular Sciences · doi:10.3390/ijms23094591 · Cited by 173 · see citing works
Mining human microbiomes reveals an untapped source of peptide antibiotics
Torres, M. D. T., Brooks, E. F., Cesaro, A., Sberro, H., Gill, M. O., Nicolaou, C., Bhatt, A. S., & de la Fuente‐Núñez, C. (2024) · Cell · doi:10.1016/j.cell.2024.07.027 · Cited by 172 · see citing works
The Virtual Lab of AI agents designs new SARS-CoV-2 nanobodies
Swanson, K., Wu, W., Bulaong, N. L., Pak, J. E., & Zou, J. (2025) · Nature · doi:10.1038/s41586-025-09442-9 · Cited by 169 · see citing works
Mechanisms of calcium homeostasis orchestrate plant growth and immunity
Wang, C., Tang, R., Kou, S., Xu, X., Lu, Y., Rauscher, K., Voelker, A., & Luan, S. (2024) · Nature · doi:10.1038/s41586-024-07100-0 · Cited by 169 · see citing works
Direct observation of motor protein stepping in living cells using MINFLUX
Deguchi, T., Iwanski, M. K., Schentarra, E.-M., Heidebrecht, C., Schmidt, L., Heck, J., Weihs, T., Schnorrenberg, S., Hoess, P., Liu, S., et al. (2023) · Science · doi:10.1126/science.ade2676 · Cited by 167 · see citing works
Easy and accurate protein structure prediction using ColabFold
Kim, G., Lee, S., Karin, E. L., Kim, H., Moriwaki, Y., Ovchinnikov, S., Steinegger, M., & Mirdita, M. (2024) · Nature Protocols · doi:10.1038/s41596-024-01060-5 · Cited by 167 · see citing works
Self-citationAFsample: improving multimer prediction with AlphaFold using massive sampling
Wallner, B. (2023) · Bioinformatics · doi:10.1093/bioinformatics/btad573 · Cited by 164 · see citing works
Navigating the landscape of enzyme design: from molecular simulations to machine learning.
Zhou, J., & Huang, M.-L. (2024) · Chemical Society Reviews · doi:10.1039/d4cs00196f · Cited by 106 · see citing works
Cyclic peptide structure prediction and design using AlphaFold2
Rettie, S., Campbell, K. V., Bera, A. K., Kang, A., Kozlov, S., Bueso, Y. F., Cruz, J. D. L., Ahlrichs, M., Cheng, S., Gerben, S., et al. (2025) · Nature Communications · doi:10.1038/s41467-025-59940-7 · Cited by 164 · see citing works
Self-citationAlphaPulldown—a python package for protein–protein interaction screens using AlphaFold-Multimer
Yu, D., Chojnowski, G., Rosenthal, M., & Kosinski, J. (2023) · Bioinformatics · doi:10.1093/bioinformatics/btac749 · Cited by 162 · see citing works
PEP-FOLD4: a pH-dependent force field for peptide structure prediction in aqueous solution
Rey, J., Murail, S., de Vries, S. J., Derreumaux, P., & Tufféry, P. (2023) · Nucleic Acids Research · doi:10.1093/nar/gkad376 · Cited by 162 · see citing works
PeSTo: parameter-free geometric deep learning for accurate prediction of protein binding interfaces
Krapp, L. F., Abriata, L. A., Rodríguez, F. C., & Peraro, M. D. (2023) · Nature Communications · doi:10.1038/s41467-023-37701-8 · Cited by 159 · see citing works
Axonemal structures reveal mechanoregulatory and disease mechanisms
Walton, T., Gui, M., Velkova, S. A., Fassad, M. R., Hirst, R. A., Haarman, E. G., O’Callaghan, C., Bottier, M., Burgoyne, T., Mitchison, H. M., et al. (2023) · Nature · doi:10.1038/s41586-023-06140-2 · Cited by 160 · see citing works
Phyre2.2: A Community Resource for Template-based Protein Structure Prediction
Powell, H. R., Islam, S. A., David, A., & Sternberg, M. J. E. (2025) · Journal of Molecular Biology · doi:10.1016/j.jmb.2025.168960 · Cited by 159 · see citing works
AlphaFold predictions of fold-switched conformations are driven by structure memorization
Chakravarty, D., Schafer, J. W., Chen, E. A., Thole, J. F., Ronish, L. A., Lee, M., & Porter, L. L. (2024) · Nature Communications · doi:10.1038/s41467-024-51801-z · Cited by 159 · see citing works
Aberrant phase separation and nucleolar dysfunction in rare genetic diseases
Mensah, M. A., Niskanen, H., Magalhães, A. P., Basu, S., Kircher, M., Sczakiel, H. L., Reiter, A. M. V., Elsner, J., Meinecke, P., Biskup, S., et al. (2023) · Nature · doi:10.1038/s41586-022-05682-1 · Cited by 156 · see citing works
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
“For targets without known homologs, such as computationally designed proteins, increasing the Perspective https://doi.org/10.1038/s41589-024-01638-w number of recycling iterations can improve the quality of the prediction 8 .”
BackgroundInfluentialMultiscale photocatalytic proximity labeling reveals cell surface neighbors on and between cells
Lin, Z., Schaefer, K., Lui, I., Yao, Z. W., Fossati, A., Swaney, D. L., Palar, A., Šali, A., & Wells, J. A. (2024) · Science · doi:10.1126/science.adl5763 · Cited by 155 · see citing works
Protenix - Advancing Structure Prediction Through a Comprehensive AlphaFold3 ReproductionPreprint
ByteDance AML AI4Science Team, Chen, X., Zhang, Y., Lu, C., Ma, W., Guan, J., Gong, C., Yang, J., Zhang, H., Zhang, K., et al. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.01.08.631967 · Cited by 150 · see citing works
Kryshtafovych, A., Schwede, T., Topf, M., Fidelis, K., & Moult, J. (2023) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.26617 · Cited by 148 · see citing works
“But strikingly, two versions of standard AF2 procedures (dashed black lines), one models from the ColabFold server 22 and one using a local installation of AF2 with default parameters 23 have substantially worse performance, and performance does fall-off markedly as homology information weakens.”
MethodMaking it Rain: Cloud-Based Molecular Simulations for Everyone
Arantes, P. R., Polêto, M. D., Pedebos, C., & Ligabue-Braun, R. (2021) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.1c00998 · Cited by 148 · see citing works
Protein complexes in cells by AI‐assisted structural proteomics
(2023) · doi:10.15252/msb.202311544 · Cited by 147 · see citing works
Protein remote homology detection and structural alignment using deep learning
Hamamsy, T., Morton, J. T., Blackwell, R., Berenberg, D., Carriero, N. J., Gligorijević, V., Strauss, C. E. M., Leman, J. K., Cho, K., & Bonneau, R. (2023) · Nature Biotechnology · doi:10.1038/s41587-023-01917-2 · Cited by 145 · see citing works
Yin, R., & Pierce, B. G. (2024) · Protein Science · doi:10.1002/pro.4865 · Cited by 145 · see citing works
An atlas of protein homo-oligomerization across domains of life
Schweke, H., Pacesa, M., Levin, T., Goverde, C. A., Kumar, P., Duhoo, Y., Dornfeld, L. J., Dubreuil, B., Georgeon, S., Ovchinnikov, S., et al. (2024) · Cell · doi:10.1016/j.cell.2024.01.022 · Cited by 144 · see citing works
Self-citationKoonin, E. V., Dolja, V. V., & Krupovìč, M. (2022) · Cell Host & Microbe · doi:10.1016/j.chom.2022.06.008 · Cited by 143 · see citing works
NCOA4 drives ferritin phase separation to facilitate macroferritinophagy and microferritinophagy
Ohshima, T., Yamamoto, H., Sakamaki, Y., Saito, C., & Mizushima, N. (2022) · The Journal of Cell Biology · doi:10.1083/jcb.202203102 · Cited by 143 · see citing works
Santana, D. J., Anku, J. A. E., Zhao, G., Zarnowski, R., Johnson, C. J., Hautau, H., Visser, N. D., Ibrahim, A. S., Andes, D., Nett, J. E., et al. (2023) · Science · doi:10.1126/science.adf8972 · Cited by 142 · see citing works
Protein language models learn evolutionary statistics of interacting sequence motifs
Zhang, Z., Wayment-Steele, H. K., Brixi, G., Wang, H., Kern, D., & Ovchinnikov, S. (2024) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2406285121 · Cited by 140 · see citing works
Self-citationExploring structural diversity across the protein universe with The Encyclopedia of Domains
Lau, A. M., Bordin, N., Kandathil, S. M., Sillitoe, I., Waman, V. P., Wells, J., Orengo, C. A., & Jones, D. T. (2024) · Science · doi:10.1126/science.adq4946 · Cited by 140 · see citing works
Improved AlphaFold modeling with implicit experimental information
Terwilliger, T. C., Poon, B. K., Afonine, P. V., Schlicksup, C. J., Croll, T. I., Millán, C., Richardson, J. S., Read, R. J., & Adams, P. D. (2022) · Nature Methods · doi:10.1038/s41592-022-01645-6 · Cited by 140 · see citing works
Assessment of three‐dimensional RNA structure prediction in CASP15
Das, R., Kretsch, R. C., Simpkin, A. J., Mulvaney, T., Pham, P., Rangan, R., Bu, F., Keegan, R. M., Topf, M., Rigden, D. J., et al. (2023) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.26602 · Cited by 139 · see citing works
- 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 · 2016: 1 · 2015: 1 · 2013: 2 · 2012: 2
- 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
- Physical Sciences › Materials Science › Materials Chemistry › 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.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.
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
