CoChat
Sign Up
PreprintCited by 5,542Protein Structure and DynamicsMachine Learning in BioinformaticsEnzyme Structure and Function

Accurate prediction of protein structures and interactions using a 3-track network

Open access
Baek, Minkyung, DiMaio, Frank, Anishchenko, Ivan, Dauparas, Justas, Ovchinnikov, Sergey, Lee, Gyu Rie, Wang, Jue, Cong, Qian +24 more·2021·Science
DOI 10.1101/2021.06.14.448402PMC PMC7612213
Open this paper in CoChat — free account

Abstract

DeepMind presented remarkably accurate protein structure predictions at the CASP14 conference. We explored network architectures incorporating related ideas and obtained the best performance with a 3-track network in which information at the 1D sequence level, the 2D distance map level, and the 3D coordinate level is successively transformed and integrated. The 3-track network produces structure predictions with accuracies approaching those of DeepMind in CASP14, enables rapid solution of challenging X-ray crystallography and cryo-EM structure modeling problems, and provides insights into the functions of proteins of currently unknown structure. The network also enables rapid generation of accurate models of protein-protein complexes from sequence information alone, short circuiting traditional approaches which require modeling of individual subunits followed by docking. We make the method available to the scientific community to speed biological research. One-Sentence Summary: Accurate protein structure modeling enables rapid solution of structure determination problems and provides insights into biological function.

AI snapshot

Written from the paper’s full text.

TL;DR

RoseTTAFold is a 3-track neural network that integrates 1D sequence, 2D distance-map, and 3D coordinate information to predict protein structures and interactions. It approaches AlphaFold2 accuracy in CASP14, solves challenging crystallography and cryo-EM modeling problems, and generates protein-complex models directly from sequence.

Key findings

  • In CASP14, the 3-track RoseTTAFold outperformed the top 2 server groups (Zhang-server and BAKER-ROSETTASERVER), the BAKER human group, and the authors’ 2-track models, though its performance was still not as good as AlphaFold2.
  • On CAMEO, over 69 medium and hard targets released May 15–June 19, 2021, RoseTTAFold outperformed all other evaluated servers, including Robetta, IntFold6-TS, BestSingleTemplate, and SWISS-MODEL.
  • RoseTTAFold models enabled molecular-replacement solutions for four previously unsolved crystallographic datasets; for the cryo-EM p101 Gβγ binding domain, the Cα-RMSD between predicted and refined structure was 3.0 Å over the beta-sheets.
  • Among 693 domains from human disease-related proteins, over one-third had predicted lDDT > 0.8, corresponding to an average Cα-RMSD of 2.6 Å on CASP14 targets; many protein-complex models had TM-score > 0.8, with complex generation taking ~30 min on a 24G TITAN RTX GPU.

Methods

  • The 3-track network transforms and integrates information at the 1D sequence/MSA level, the 2D distance/orientation-map level, and the 3D backbone-coordinate level, with information flowing between tracks; training used discontinuous sequence crops spanning 260 residues.
  • Final 3D structures were generated either by feeding predicted distance/orientation distributions into pyRosetta for all-atom models or by an end-to-end SE(3)-equivariant layer producing backbone coordinates directly.
  • Evaluation included CASP14 targets, CAMEO blind server assessment, four X-ray molecular-replacement datasets, a cryo-EM PI3Kγ complex, GPCRs, 693 human disease-related domains, and paired-sequence complexes of known structure, using accuracy comparisons, Cα-RMSD, lDDT, TM-score, and MR/cryo-EM fit.

Limitations

  • Stated: hardware memory limitations constrained model size and training; performance remained below AlphaFold2; the end-to-end version was less accurate than the pyRosetta version due to incomplete optimization and neglect of side-chain information; the accuracy predictor did not consistently identify models better than the rapid single-pass method; the network was trained on monomeric proteins, not complexes; and current RoseTTAFold uses only the top 1000 sequences due to memory limits.
  • Stated: the Perceiver-based extension for using all available sequences gave promising initial results, but more training is required for rigorous comparison.
  • Evident: CASP14 and CAMEO evaluations cover limited target sets and time windows, and the end-to-end model does not produce side chains.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1101/2021.06.14.448402

Cites

85 of 85 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

13 of 36 cited sources checked · 0 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Deep learning methods such as AlphaFold and trRosetta, which extract information from the PDB, outperform traditional approaches that explicitly model the folding process.

    “The bi-annual Critical Assessment of Structure (CASP) meetings have demonstrated that deep learning methods such as AlphaFold (1, 2) and trRosetta (3), that extract information from the large database of known protein structures in the PDB, outperform more traditional approaches that explicitly model the folding process.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [1] John Jumper , Richard Evans , Alexander Pritzel , Tim Green , Michael Figurnov , Kathryn Tunyasuvunakool , Olaf Ronneberger , Russ Bates , Augustin Žídek , Alex Bridgland , Clemens Meyer , Simon A A Kohl , Anna Potapenko , Andrew J Ballard , Andrew Cowie , Bernardino Romera-Paredes , Stanislav Nikolov , Rishub Jain , Jonas Adler , Trevor Back , Stig Petersen , David Reiman , Martin Steinegger , Michalina Pacholska , David Silver , Oriol Vinyals , Andrew W Senior , Koray Kavukcuoglu , Pushmeet Kohli , Demis Hassabis ., in Fourteenth Critical Assessment of Techniques for Protein Structure Predic

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [2] source

      The Introduction (page not provided) says deep networks improved protein-structure prediction accuracy, notes an AlphaFold observation at CASP13, and reports that trRosetta outperformed other methods in benchmark tests. The Conclusion says the network learned from native protein structures and contrasts it with physically based approaches. The paper does not establish the claimed CASP comparison in which both AlphaFold and trRosetta outperform approaches that explicitly model folding. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
    • cites [3] source
  • Findingcitation

    The RoseTTAFold CERS1 model for residues 98 to 304 (Pfam TLC domain) includes six transmembrane helices that traverse the membrane in an up and down arrangement.

    “The RoseTTAFold CERS1 model for residues 98 to 304 (Pfam TLC domain) (29) includes six TMH that traverse the membrane in an up and down arrangement (Fig. 3E).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [29] source

      The paper’s Abstract (page unavailable) describes coevolution-based protein interaction models in E. coli and M. tuberculosis, not a CERS1 model. Searches for CERS1, the Pfam TLC domain, RoseTTAFold, and six transmembrane helices found no passage in our copy addressing the cited domain identification or the stated membrane arrangement. (LLM, conf=0.76)

      Experimental — accuracy not yet measured
  • Findingown data

    For SLP, only a distant model covering part of the N-terminal domain (38% of the sequence) was available in the PDB.

    “in the case of SLP, only a distant model covering part of the N-terminal domain (38% of the sequence) was available in the PDB”
    Quote in the paperSubstantial claim
  • Findingown data

    trRosetta failed to predict the correct fold for the p101 Gβγ binding domain with the same MSA input.

    “trRosetta failed to predict the correct fold with the same MSA input (fig. S6)”
    Quote in the paperSubstantial claim
  • Findingown data

    The 3-track models with attention at the 1D, 2D, and 3D levels clearly outperformed the top 2 server groups, the BAKER human group, and the 2-track attention models on CASP14 targets.

    “The 3-track models with attention operating at the 1D, 2D, and 3D levels and information flowing between the three levels were the best models we tested (Fig. 1B), clearly outperforming the top 2 server groups (Zhang-server and BAKER-ROSETTASERVER), BAKER human group (ranked second among all groups), and our 2-track attention models on CASP14 targets.”
    Quote in the paperSubstantial claim
  • May combine two claims. 24
  • Classified as probably not a substantial claim. 13
  • The quote only partly states this claim. 2

  • May combine two claims.

    Findingown data

    RoseTTAFold has been evaluated since May 15th, 2021 on CAMEO; over the 69 medium and hard targets released during this time, it outperformed all other servers evaluated including Robetta, IntFold6-TS, BestSingleTemplate, and SWISS-MODEL.

    “RoseTTAFold has been evaluated since May 15th, 2021 on CAMEO; over the 69 medium and hard targets released during this time (May 15th, 2021 ~ June 19th, 2021), it outperformed all other servers evaluated in the experiment including Robetta (3), IntFold6-TS (8), BestSingleTemplate (9), and SWISS-MODEL (10) (Fig. 1C).”
    Quote in the paperSubstantial claim3 of 4 sources checked
    • cites [3] source
    • cites [8] 2021, GPCRdb in 2021: integrating GPCR sequence, structure and function, Nucleic Acids Res source

      Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [9] 2020, Improving homology modeling from low-sequence identity templates in Rosetta: A case study in GPCRs, PLoS Comput. Biol source

      Split into 2 atoms — 1 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [10] 2021, Improved protein structure refinement guided by deep learning based accuracy estimation, Nat. Commun source

      Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    The Continuous Automated Model Evaluation (CAMEO) experiment tests structure prediction servers blindly on protein structures as they are submitted to the PDB.

    “Fortunately, the Continuous Automated Model Evaluation (CAMEO) experiment (7) tests structure prediction servers blindly on protein structures as they are submitted to the PDB.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [7] source

      The Introduction discusses tools for finding protein-structure templates, and the Results’ Sensitivity benchmark describes tests using SCOP domains. Neither passage describes CAMEO or blind testing of prediction servers as structures are submitted to the PDB. Searches for CAMEO and for blind testing with PDB submissions found no passage addressing the claim. (LLM, conf=0.72)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    Based on the model, known disease-causing mutations in TANGO2 could act by hindering catalysis (R26K, R32Q, and L50P, near active site) or produce steric clashes (G154R) in the hydrophobic core.

    “Based on the model, known mutations that cause disease (magenta spheres in Fig. 3A) could act by hindering catalysis (R26K, R32Q, and L50P, near active site) or produce steric clashes (G154R) (19) in the hydrophobic core.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [19] source

      The Structural analysis section describes a crystal model of VAP1, not a model of TANGO2. Searches for TANGO2, G154R, R26K, R32Q, and L50P found no passage addressing those mutations or the proposed effects on catalysis and the hydrophobic core. (LLM, conf=0.77)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Deficiencies in TANGO2 lead to metabolic disorders, and the protein plays an unknown role in Golgi membrane redistribution into the ER.

    “Deficiencies in TANGO2 (transport and Golgi organization protein 2) lead to metabolic disorders, and the protein plays an unknown role in Golgi membrane redistribution into the ER (16, 17).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [16] source
    • cites [17] 2019, A Disintegrin and Metalloproteinase (ADAM) and ADAM with thrombospondin motifs (ADAMTS) family in vascular biology and disease, Biochem. Pharmacol source

      The ADAMTS Functions section discusses deficiencies and disorders involving ADAMTS proteins, not TANGO2. Searches for TANGO2 and for Golgi membrane redistribution into the ER found no passage in our copy addressing either part of the claim. (LLM, conf=0.77)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The ADAM and ADAMTS families of metalloproteases are encoded by over 40 human genes, mediate cell-cell and cell-matrix interactions, and are involved in a range of human diseases.

    “The ADAM (A Disintegrin And Metalloprotease) and ADAMTS families of metalloproteases are encoded by over 40 human genes, mediate cell-cell and cell-matrix interactions (20, 21) and are involved in a range of human diseases, including cancer metastasis, inflammatory disorders, neurological diseases and asthma (21, 22).”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [20] source
    • cites [21] source
    • cites [22] source

      The Introduction (page unavailable) identifies ADAM proteins as metalloproteases and discusses ADAM10 in relation to Alzheimer’s disease, supporting the neurological-disease example. Our copy does not state that the ADAM and ADAMTS families comprise over 40 human genes, mediate cell–cell or cell–matrix interactions, or are involved in the other listed diseases. Searches for those family-wide claims found no supporting passage. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The ADAMs contain prodomain and metalloprotease domains; the fold of the metalloprotease is known, but not that of the prodomain, which has no homologs of known structure.

    “The ADAMs contain prodomain and metalloprotease domains; the fold of the metalloprotease is known (23, 24), but not that of the prodomain, which has no homologs of known structure.”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [23] source
    • cites [24] source

      The Structural features section (page unavailable) describes ceramide synthase domains, not ADAMs. Searches for ADAM prodomain and metalloprotease folds found no passage addressing the claim in our copy. (LLM, conf=0.78)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    Benchmark tests on GPCR sequences with determined structures showed that RoseTTAFold models for both active and inactive states can be quite accurate even in the absence of close homologs with known structures, and better than those in current GPCR model databases.

    “Benchmark tests on GPCR sequences with determined structures showed that RoseTTAFold models for both active and inactive states can be quite accurate even in the absence of close homologs with known structures (and better than those in current GPCR model databases (14, 15); fig. S7)”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [14] 2016, Recurrent Muscle Weakness with Rhabdomyolysis, Metabolic Crises, and Cardiac Arrhythmia Due to Bi-allelic TANGO2 Mutations, Am. J. Hum. Genet source
    • cites [15] source

      The cited paper’s Conclusions (p. 277) discuss the ADAM protein family and its possible functions. Searches of our copy for RoseTTAFold, GPCR structural models, and model databases found no passage addressing the benchmark or comparison claimed here. (LLM, conf=0.78)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    A previously published cryo-EM map of the IL-12 receptor complex indicated a similar topology to that of the IL-23 receptor, but its resolution was not sufficient to observe the detailed interaction between IL-12Rβ2 and IL-12p35.

    “A previously published cryo-EM map of the IL-12 receptor complex indicated a similar topology to that of the IL-23 receptor; however, the resolution was not sufficient to observe the detailed interaction between IL-12Rβ2 and IL-12p35 (34).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [34] A. Vaswani , N. Shazeer , N. Parmar , J. Uszkoreit , L. Jones , A. N. Gomez , L. Kaiser , I. Polosukhin , Attention Is All You Need. arXiv [cs.CL] (2017), (available at http://arxiv.org/abs/1706.03762). source

      The cited paper’s Background (p. 2) discusses self-attention, and its Model Variations section (p. 9) discusses Transformer settings. Searches for the IL-12 receptor, IL-23 receptor, cryo-EM, and the named receptor interactions found no passage addressing the claim. (LLM, conf=0.78)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingcitation

    A central crevice extends into the membrane and is lined with residues required for activity (His182 and Asp213) or conserved (W298), as well as a pathogenic mutation (H183Q) found in progressive myoclonus epilepsy and dementia that decreases ceramide levels.

    “A central crevice extends into the membrane and is lined with residues required for activity (His182 and Asp213) (30) or conserved (W298), as well as a pathogenic mutation (H183Q) found in progressive myoclonus epilepsy and dementia that decreases ceramide levels (31).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [30] 2021, Structural basis for IL-12 and IL-23 receptor sharing reveals a gateway for shaping actions on T versus NK cells, Cell source

      Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [31] source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Following sequence and template search (~1.5 hours), the end-to-end version of RoseTTAFold requires ~10 minutes on an RTX2080 GPU to generate backbone coordinates for proteins with less than 400 residues.

    “following sequence and template search (~1.5 hours), the end-to-end version of RoseTTAFold requires ~10 minutes on an RTX2080 GPU to generate backbone coordinates for proteins with less than 400 residues”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingnone given

    The pyRosetta version requires 5 minutes for network calculations on a single RTX2080 GPU and an hour for all-atom structure generation with 15 CPU cores.

    “the pyRosetta version requires 5 minutes for network calculations on a single RTX2080 GPU and an hour for all-atom structure generation with 15 CPU cores”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Over one-third of the 693 RoseTTAFold models for human disease-related protein domains have a predicted lDDT > 0.8, which corresponded to an average Cα-RMSD of 2.6 Å on CASP14 targets.

    “Over one-third of these models have a predicted lDDT > 0.8, which corresponded to an average Cα-RMSD of 2.6 Å on CASP14 targets (fig. S8).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The increased prediction accuracy was critical for success in all four molecular replacement cases, as models made with trRosetta did not yield MR solutions.

    “The increased prediction accuracy was critical for success in all cases, as models made with trRosetta did not yield MR solutions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    No homologs of the C-terminal domain of SLP or any portion of Lrbp could be detected using HHsearch.

    “while no homologs of the C-terminal domain of SLP or any portion of Lrbp could be detected using HHsearch (13)”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [13] 2021, The phenotype associated with variants in TANGO2 may be explained by a dual role of the protein in ER-to-Golgi transport and at the mitochondria, J. Inherit. Metab. Dis source

  • May combine two claims.

    Findingown data

    The top HHsearch hit for the p101 Gβγ binding domain has a statistically insignificant E-value of 40 and only covers 14 residues out of 167 residues.

    “The top HHsearch hit has a statistically insignificant E-value of 40 and only covers 14 residues out of 167 residues.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The RoseTTAFold model of TANGO2 adopts an N-terminal nucleophile aminohydrolase (Ntn) fold with well-aligned active site residues that are conserved in TANGO2 orthologs.

    “The RoseTTAFold model of TANGO2 adopts an N-terminal nucleophile aminohydrolase (Ntn) fold (Fig. 3A) with well-aligned active site residues that are conserved in TANGO2 orthologs (Fig. 3B).”
    Quote in the paperSubstantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    Approaches using iterative feeding of predicted structures as templates and random subsampling from MSAs generated ensembles containing higher accuracy models.

    “These approaches generated ensembles containing higher accuracy models”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S1 (asserted), S4 (based).

    Interpretationnone given

    The improved accuracy from combining multiple discontinuous crops arises from selecting the most relevant sequences for each region from the large number of aligned sequences.

    “We hypothesized that this arises from selecting the most relevant sequences for each region from the very large number of aligned sequences often available (fig. S4B).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    AlphaFold2 used a two-track network architecture in which information at the 1D sequence level and the 2D distance map level is iteratively transformed and passed back and forth.

    “use of a two-track network architecture in which information at the 1D sequence level and the 2D distance map level is iteratively transformed and passed back and forth”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites CASP14 conference

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    Incomplete optimization due to computer memory limitations and neglect of side chain information likely explain the poorer performance of the end-to-end version compared to the pyRosetta version.

    “Incomplete optimization due to computer memory limitations and neglect of side chain information likely explain the poorer performance of the end-to-end version compared to the pyRosetta version (Fig. 1B; the latter incorporates side chain information at the all-atom relaxation stage)”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The relatively low compute cost makes it straightforward to incorporate the methods in a public server and predict structures for large sets of proteins.

    “The relatively low compute cost makes it straightforward to incorporate the methods in a public server and predict structures for large sets of proteins, for example, all human GPCRs, as described below.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    AlphaFold2 started from multiple sequence alignments rather than from more processed features such as inverse covariance matrices derived from MSAs.

    “starting from multiple sequence alignments (MSAs) rather than from more processed features such as inverse covariance matrices derived from MSAs”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites CASP14 conference

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    Since SE(3)-equivariant layers are used in the main body of the 3-track model, the added gain from the final SE(3) layer is likely less than in the AlphaFold2 case.

    “since SE(3)-equivariant layers are used in the main body of the 3-track model, the added gain from the final SE(3) layer is likely less than in the AlphaFold2 case”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    There is a cysteine in an extension following the predicted ADAM33 prodomain barrel; taken together, these data are consistent with experimental data suggesting that the ADAM prodomain inhibits metalloprotease activity using a cysteine switch.

    “There is a cysteine in an extension following the predicted prodomain barrel; taken together, these data are consistent with experimental data suggesting that the ADAM prodomain inhibits metalloprotease activity using a cysteine switch (26).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [26] source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    The poorer performance of the 3-track model compared to AlphaFold2 could reflect hardware limitations, alternative architectures or loss formulations, or more intensive use of the network for inference.

    “This could reflect hardware limitations that limited the size of the models we could explore, alternative architectures or loss formulations, or more intensive use of the network for inference.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Structure information is not available for any of the CerS enzymes or their homologs, and the number and orientation of transmembrane helices are not known.

    “Structure information is not available for any of the CerS enzymes or their homologs, and the number and orientation of transmembrane helices (TMH) are not known (28).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [28] source

  • May combine two claims.

    Interpretationreasoning

    Conserved residues within ADAM33 orthologs line one side of the barrel and likely interact with the metalloprotease.

    “Conserved residues within ADAM33 orthologs line one side of the barrel and likely interact with the metalloprotease (Fig. 3D).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The RoseTTAFold model suggests a role for the IL-12Rβ2 N-terminal peptide (residue 24-31) in IL-12 binding, with IL-12Rβ2 D26 possibly interacting with nearby K190 and K194 in IL-12p35.

    “In addition, the model suggests a role for the IL-12Rβ2 N-terminal peptide (residue 24-31) in IL-12 binding not observed in the IL-12 cryo-electron microscopy (IL-12Rβ2 D26 may interact with nearby K190 and K194 in IL-12p35), which may provide an avenue to target the interaction between IL-12 and IL-12Rβ2 specifically.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Transmembrane spanning Ceramide synthase (CERS1) is a key enzyme in sphingolipid metabolism which uses acyl-CoA to generate ceramides with various acyl chain lengths that regulate differentiation, proliferation, and apoptosis.

    “Transmembrane spanning Ceramide synthase (CERS1) is a key enzyme in sphingolipid metabolism which uses acyl-CoA to generate ceramides with various acyl chain lengths that regulate differentiation, proliferation, and apoptosis (27).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [27] source

  • May combine two claims. The quote only partly states this claim.

    Interpretationreasoning

    RoseTTAFold enables direct building of accurate models of protein-protein complexes from sequence information, rapidly and without the standard procedure of building individual subunit models and then rigid-body docking.

    “Thus, the network enables the direct building of structure models for protein-protein complexes from sequence information, short circuiting the standard procedure of building models for individual subunits and then carrying out rigid-body docking.”
    • Also stated in: § Conclusions · plain
      “RoseTTAFold enables solutions of challenging X-ray crystallography and cryo-EM modeling problems, provides insight into protein function in the absence of experimentally determined structures, and rapidly generates accurate models of protein-protein complexes.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductnone given

    The network was trained on monomeric proteins, not complexes, so there may be some training set bias in the monomer structures, but there is none for the complexes.

    “The network was trained on monomeric proteins, not complexes, so there may be some training set bias in the monomer structures, but there is none for the complexes.”
    Quote in the paperNot a substantial claim
31 of 31

  • The paper relies on this without support.

    The prediction of protein structure from amino acid sequence information alone has been a longstanding challenge.

    “The prediction of protein structure from amino acid sequence information alone has been a longstanding challenge.”

  • The paper relies on this without support.

    Solution of X-ray structures by molecular replacement often requires quite accurate models.

    “Solution of X-ray structures by molecular replacement (MR) often requires quite accurate models.”

  • The paper relies on this without support.

    Building atomic models of protein assemblies from cryo-EM maps can be challenging in the absence of homologs with known structures.

    “Building atomic models of protein assemblies from cryo-EM maps can be challenging in the absence of homologs with known structures.”

  • The paper relies on this without support.

    Experimental structure determination can provide considerable insight into biological function and mechanism.

    “Experimental structure determination can provide considerable insight into biological function and mechanism.”

  • The paper relies on this without support.

    Protein structures can provide insight into how mutations in key proteins lead to human disease.

    “Protein structures can provide insight into how mutations in key proteins lead to human disease.”

  • The paper relies on this without support.

    Blind structure prediction tests are needed to assess any new protein structure prediction method.

    “Blind structure prediction tests are needed to assess any new protein structure prediction method”
6 of 6

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?

  • RF has properties well suited for use in a protein design DDPM: it generates protein structures with high precision, operates on a rigid-frame representation of residues with rotational equivariance, and has an architecture enabling conditioning on design specifications at residue, inter-residue distance and orientation, and 3D coordinate levels.

    “RF has properties well suited for use in a protein design DDPM (Fig. 1a): it generates protein structures with high precision, operates on a rigid-frame representation of residues with rotational equivariance and has an architecture enabling conditioning on design specifications at the individual residue, inter-residue distance and orientation, and 3D coordinate levels.”

    From De novo design of protein structure and function with RFdiffusion (2023) · cites this paper as [18]

    Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

    Experimental — accuracy not yet measured

  • AI modeling paradigms have been used successfully in the biomolecular realm, for example, to predict 3D molecular structures from sequences and interactions between different molecular components.

    “Such modeling paradigms have been used successfully in the biomolecular realm, for example, to predict three-dimensional (3D) molecular structures from sequences32–34 and interactions between different molecular components.35–38”

    From How to build the virtual cell with artificial intelligence: Priorities and opportunities (2024) · cites this paper as [33]

  • Predicted structures for gpsB/0353 from AlphaFold2 and RoseTTAFold align nearly exactly with the crystal structure of the GpsB protein from B. subtilis (RCSB PDB: 4UG3) when aligned at the N-terminal coiled-coil domain using STAMP.

    “Predicted structures for gene gpsB/0353 from AlphaFold222 (and the related protein structure prediction software RoseTTAFold36) align nearly exactly with the crystal structure of the GpsB protein from B. subtilis (RCSB PDB: 4UG3(58)) when aligned only at the N-terminal coiled-coil domain using STAMP41 and visualizing in VMD42 as is shown in Figure 1A.”

    From Toward the Complete Functional Characterization of a Minimal Bacterial Proteome (2022) · cites this paper as [36]

  • The RoseTTAFold pipeline for complex modelling only generates MSAs for bacterial protein complexes, while the proteins in the test set are mainly Eukaryotic.

    “The RoseTTAFold pipeline for complex modelling only generates MSAs for bacterial protein complexes, while the proteins in our test set are mainly Eukaryotic.”

    From Improved prediction of protein-protein interactions using AlphaFold2 (2021) · cites this paper as [17]

    The ‘Direct generation of protein-protein complex models’ section says RoseTTAFold was tested on paired sequence alignments and describes a model of a human complex, but it does not say that its pipeline generates MSAs only for bacterial complexes. The source also cannot establish the composition of the citing authors’ test set. Searches for bacterial-only MSA generation and eukaryotic complexes did not resolve either point. (LLM, conf=0.72)

    Experimental — accuracy not yet measured

  • RosettaFold performed better than trRosetta in guiding protein design by functional-site-constrained hallucination.

    “Following the development of RosettaFold (RF) (16) we found that it performed better than trRosetta in guiding protein design by functional-site-constrained hallucination (Fig. S1G)”

    From Scaffolding protein functional sites using deep learning (2022) · cites this paper as [16]

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. 2011, phenix. ensembler: a tool for multiple superposition, Comput Crystallogr Newsl, 2, 8
  2. 2015, in ACTA CRYSTALLOGRAPHICA A-FOUNDATION AND ADVANCES (INT UNION CRYSTALLOGRAPHY 2 ABBEY SQ, CHESTER, CH1 2HU, ENGLAND, 71, S19

Cited by

Sorted by citations

Related

Authors
Baek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee, G. R., Wang, J., Cong, Q., Kinch, L. N., Schaeffer, R. D., Millán, C., Park, H., Adams, C., Glassman, C. R., DeGiovanni, A., Pereira, J. H., Rodrigues, A. V., van Dijk, A. A., Ebrecht, A. C., Opperman, D. J., Sagmeister, T., Buhlheller, C., Pavkov-Keller, T., Rathinaswamy, M. K., Dalwadi, U., Yip, C. K., Burke, J. E., Garcia, K. C., Grishin, N. V., Adams, P. D., Read, R. J., & Baker, D.
Affiliations
  • University of Washington (US)
  • Harvard University (US)
  • The University of Texas Southwestern Medical Center (US)
  • Howard Hughes Medical Institute (US)
  • University of Cambridge (GB)
  • Stanford Medicine (US)
  • Lawrence Berkeley National Laboratory (US)
  • North-West University (ZA)
  • University of the Free State (ZA)
  • University of Graz (AT)
  • University of Victoria (CA)
  • University of British Columbia (CA)
Corresponding author
Baker, David
ORCID
Venue
Science
Publisher
openRxiv
Published
2021-06-15
Volume
373
Issue
6557
Pages
871-876
ISSN
0036-8075 (ISSN-L)
DOI
10.1101/2021.06.14.448402
PMC
PMC7612213
OpenAlex
W3186179742
Semantic Scholar
51362860fb8ad08255660303550ed4bc345535fc
Cited by
5,542
References
85
Citations by year
2026: 753 · 2025: 1082 · 2024: 1203 · 2023: 1196 · 2022: 1039 · 2021: 257 · 2020: 3 · 2013: 2 · 2012: 4
FWCI
249.46
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Science (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • eScholarship (California Digital Library) (submittedVersion, cc-by)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion, cc-by)
  • PubMed Central (submittedVersion, cc-by)
  • Desy Publications Database (Deutsches Elektronen-Synchrotron DESY) (submittedVersion)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • Apollo (cc-by)
Funders
  • National Science Foundation (GM127390, 1937533)
  • Welch Foundation (I‐1505)
  • Washington Research Foundation
  • Microsoft
  • TU Graz, Internationale Beziehungen und Mobilitätsprogramme
  • Wellcome Trust (Z/17/Z, 209407/Z/17/Z, 209407)
  • G. Harold and Leila Y. Mathers Foundation
  • Diamond Light Source
  • Open Philanthropy Project
  • Global Challenges Research Fund (ST/R002754/1)
  • Schmidt Family Foundation
  • European Synchrotron Radiation Facility
  • Austrian Science Fund (P 29432, DOC 50)
  • National Institutes of Health (5r35gm127390-05, DP5OD026389, P01GM063210, GM127390, RO1-AI51321)
  • Canadian Institutes of Health Research (168998, 168907)
  • Science and Technology Facilities Council (ST/R002754/1)
  • Deutsches Elektronen-Synchrotron
  • National Science Foundation of Sri Lanka (DBI 1937533)
  • National Institute of General Medical Sciences (R35 GM127390, R01 GM123089, P01 GM063210)
  • National Institute of Allergy and Infectious Diseases (R01 AI051321)
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-protein complex prediction, distance maps, sequence-based structure prediction, CASP14, protein function prediction
MeSH terms
Deep Learning, Amino Acid Sequence, Computer Simulation, Membrane Proteins, Models, Molecular, Protein Conformation, Proteins, Neural Networks, Computer, Protein Folding, Crystallography, X-Ray, Cryoelectron Microscopy, Protein Subunits, Databases, Protein, Receptors, G-Protein-Coupled, Multiprotein Complexes, Sphingosine N-Acyltransferase, ADAM Proteins

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.

Cited sources are checked by 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