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

The DOI matches its Crossref record; 1 of the 1 references it cites were found in a registry or an index.

Physical-aware model accuracy estimation for protein complex using deep learning method

Open access
Wang, Haodong, Sun, Meng, Xie, Lei, Liu, Dong, Zhang, Guijun·2024·Computational and Structural Biotechnology Journal
DOI 10.1101/2024.10.31.621211PMC PMC11799971

Preprint — published as 10.1016/j.csbj.2025.01.017 · Paper page

Open this paper in CoChat — free accountOpen PDF

Abstract

With the breakthrough of AlphaFold2 on monomers, the research focus of structure prediction has shifted to protein complexes, driving the continued development of new methods for multimer structure prediction. Therefore, it is crucial to accurately estimate quality scores for the multimer model independent of the used prediction methods. In this work, we propose a physical-aware deep learning method, DeepUMQA-PA, to evaluate the residue-wise quality of protein complex models. For the input complex model, the residue-based contact area and orientation features were first constructed using Voronoi tessellation, representing the potential physical interactions and hydrophobic properties. Then, the relationship between local residues and the overall complex topology as well as the inter-residue evolutionary information are characterized by geometry-based features, protein language model embedding representation, and knowledge-based statistical potential features. Finally, these features are fed into a fused network architecture employing equivalent graph neural network and ResNet network to estimate residue-wise model accuracy. Experimental results on the CASP15 test set demonstrate that our method outperforms the state-of-the-art method DeepUMQA3 by 3.69% and 3.49% on Pearson and Spearman, respectively. Notably, our method achieved 16.8% and 15.5% improvement in Pearson and Spearman, respectively, for the evaluation of nanobody-antigens. In addition, DeepUMQA-PA achieved better MAE scores than AlphaFold-Multimer and AlphaFold3 self-assessment methods on 43% and 50% of the targets, respectively. All these results suggest that physical-aware information based on the area and orientation of atom-atom and atom-solvent contacts has the potential to capture sequence-structure-quality relationships of proteins, especially in the case of flexible proteins. The DeepUMQA-PA server is freely available at http://zhanglab-bioinf.com/DeepUMQA-PA/ .

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1101/2024.10.31.621211

Cites

1 of 1 references verified0 citation errors · 0 not checkable

Statistics

Figures

Graphical abstract
Graphical abstract CC BY-NC-ND · PubMed Central
Fig. 1 The pipeline of DeepUMQA-PA. a) Data preparation. Protein complex structure and sequences were taken as input. b) Features extraction. Physical-aware contact features, geometry-based features, embedding features, and knowledge-based statistical energy features of given protein complex structure are extracted from the input structural and sequence information. c) Network architecture. The graph neural network was fused with ResNet network with attention mechanisms to estimate residue-wise prediction accuracy.
Fig. 1 The pipeline of DeepUMQA-PA. a) Data preparation. Protein complex structure and sequences were taken as input. b) Features extraction. Physical-aware contact features, geometry-based features, embedding features, and knowledge-based statistical energy features of given protein complex structure are extracted from the input structural and sequence information. c) Network architecture. The graph neural network was fused with ResNet network with attention mechanisms to estimate residue-wise prediction accuracy.CC BY-NC-ND · PubMed Central
Fig. 2 a) Schematic diagram of atomic contact between atom ai (red) and atom aj (purple). r1 and r2 are the van der Waals radius of atom ai and atom aj, d is the distance between atoms, and p is a point on the Voronoi surface, where the surface distances from p to atom ai and atom aj are both σ. b) Schematic diagram of the residue-level contact surface orientation for homodimer 1B5D. The blue arrow represents the contact surface orientation of surface residues, and the yellow arrow represents the contact surface orientation of internal residues.
Fig. 2 a) Schematic diagram of atomic contact between atom ai (red) and atom aj (purple). r1 and r2 are the van der Waals radius of atom ai and atom aj, d is the distance between atoms, and p is a point on the Voronoi surface, where the surface distances from p to atom ai and atom aj are both σ. b) Schematic diagram of the residue-level contact surface orientation for homodimer 1B5D. The blue arrow represents the contact surface orientation of surface residues, and the yellow arrow represents the contact surface orientation of internal residues.CC BY-NC-ND · PubMed Central
Fig. 3 a) Homodimer 1B5D. b) Interchain residues contact surface. Random color patches represent the contact surface of different contact residues from interchain. c) Intrachain residues contact surface. Random color patches represent the residue contact surface between the pink residue and the other contact residues (purple). Light pink patches represent the residue-solvent surface.
Fig. 3 a) Homodimer 1B5D. b) Interchain residues contact surface. Random color patches represent the contact surface of different contact residues from interchain. c) Intrachain residues contact surface. Random color patches represent the residue contact surface between the pink residue and the other contact residues (purple). Light pink patches represent the residue-solvent surface.CC BY-NC-ND · PubMed Central
Fig. 4 a The results of DeepUMQA-PA and other servers on CASP15 test set. b The results of DeepUMQA-PA and other servers on CASP15 nanobody-antigens test set.
Fig. 4 a The results of DeepUMQA-PA and other servers on CASP15 test set. b The results of DeepUMQA-PA and other servers on CASP15 nanobody-antigens test set.CC BY-NC-ND · PubMed Central
Fig. 5 a, c, e) Supplementary ablation studies of interface residue assessment accuracy on the CASP15 test set. b, d, f) Supplementary ablation studies of interface residue assessment accuracy on CASP15 nanobody-antigens test set. DeepUMQA-PA is the baseline model that uses all information. DeepUMQA-PA1 denotes a version of DeepUMQA-PA without the physical-aware contact orientation feature. DeepUMQA-PA2 denotes a version of DeepUMQA-PA without the physical-aware residue-residue contact area feature. DeepUMQA-PA3 denotes a version of DeepUMQA-PA without the physical-aware residue-solvent contact area feature.
Fig. 5 a, c, e) Supplementary ablation studies of interface residue assessment accuracy on the CASP15 test set. b, d, f) Supplementary ablation studies of interface residue assessment accuracy on CASP15 nanobody-antigens test set. DeepUMQA-PA is the baseline model that uses all information. DeepUMQA-PA1 denotes a version of DeepUMQA-PA without the physical-aware contact orientation feature. DeepUMQA-PA2 denotes a version of DeepUMQA-PA without the physical-aware residue-residue contact area feature. DeepUMQA-PA3 denotes a version of DeepUMQA-PA without the physical-aware residue-solvent contact area feature.CC BY-NC-ND · PubMed Central
Fig. 6 a) Comparison of DeepUMQA-PA and AlphaFold - Multimer self-assessment accuracy in the mean absolute error (MAE). The models of 30 targets are generated by AFM. b) Comparison of DeepUMQA-PA and AlphaFold3 self-assessment accuracy in the mean absolute error (MAE). The models of 30 targets are generated by AF3 webserver.
Fig. 6 a) Comparison of DeepUMQA-PA and AlphaFold - Multimer self-assessment accuracy in the mean absolute error (MAE). The models of 30 targets are generated by AFM. b) Comparison of DeepUMQA-PA and AlphaFold3 self-assessment accuracy in the mean absolute error (MAE). The models of 30 targets are generated by AF3 webserver.CC BY-NC-ND · PubMed Central
Fig. 7 Differences in quality estimation accuracy according to residue location in CASP15 targets. Blue represents interface residues, yellow represents surface residues, and red represents core residues.
Fig. 7 Differences in quality estimation accuracy according to residue location in CASP15 targets. Blue represents interface residues, yellow represents surface residues, and red represents core residues.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 53

  1. 10.1101/2021.10.04.463034 · DOI · Paper page
  2. 10.1002/prot.26585 · DOI
  3. 10.1093/bioinformatics/btad573 · DOI · Paper page
  4. 10.1002/prot.26570 · DOI
  5. 10.1038/S41586-024-07487-W · DOI · Paper page
  6. 10.1002/prot.26617 · DOI · Paper page
  7. 10.1002/prot.26545 · DOI
  8. 10.1002/prot.26542 · DOI
  9. 10.1002/prot.26532 · DOI
  10. 10.1002/prot.26569 · DOI · Paper page
  11. 10.1002/prot.26554 · DOI · Paper page
  12. 10.1093/bioinformatics/btad591 · DOI
  13. 10.1073/pnas.0535768100 · DOI
  14. 10.1002/(SICI)1096-987X(19970715)18:9<1113::AID-JCC1>3.0.CO;2-U · DOI
  15. 10.1002/prot.25278 · DOI
  16. 10.1038/319199a0 · DOI
  17. 2021, E(n) Equivariant Graph Neural Networks.” presented at the Proceedings of the 38th International Conference on Machine Learning, Proceedings of Machine Learning Research, 2021, 9323
  18. 10.1093/bioinformatics/btt473 · DOI · Paper page
  19. 10.1002/prot.26612 · DOI
  20. 10.1371/journal.pone.0161879 · DOI · Paper page
  21. 10.1093/bioinformatics/btae586 · DOI
  22. 10.1002/prot.24172 · DOI
  23. 10.1093/bioinformatics/btz122 · DOI
  24. 10.1038/s41467-021-21511-x · DOI · Paper page
  25. 10.1002/jcc.20681 · DOI
  26. 10.1093/bioinformatics/btac056 · DOI · Paper page
  27. 10.1002/prot.26564 · DOI
  28. 10.1093/bib/bbad420 · DOI
  29. 10.1126/science.adk8946 · DOI
  30. 10.1101/2022.07.20.500902 · DOI · Paper page
  31. 2022, Learning inverse folding from millions of predicted structures.” presented at the Proceedings of the 39th International Conference on Machine Learning, Proceedings of Machine Learning Research, 162, 8946
  32. 10.1126/science.ade2574 · DOI · Paper page
  33. 10.1016/B978-0-12-381270-4.00019-6 · DOI
  34. 10.1016/S0076-6879(04)83004-0 · DOI
  35. 10.1038/srep33509 · DOI
  36. 10.1007/978-3-319-46493-0_38 · DOI · Paper page
  37. 10.1016/j.neucom.2019.10.008 · DOI
  38. 10.1016/j.neucom.2021.03.091 · DOI · Paper page
  39. 10.48550/arXiv.1606.08415 · DOI · Paper page
  40. 10.1109/IWQoS.2018.8624183 · DOI
  41. 10.1002/pro.2007 · DOI
  42. 10.1111/opo.12636 · DOI
  43. 2022, Spearman’s correlation coefficient in statistical analysis, International Journal of Nonlinear Analysis and Applications, 13, 3249
  44. 10.1007/3-540-44886-1_25 · DOI
  45. 10.1103/RevModPhys.11.1 · DOI
  46. 10.1080/00018736100101281 · DOI
  47. 10.1038/s41467-023-37701-8 · DOI
  48. 10.1002/prot.26554 · DOI · Paper page
  49. 10.1093/bioinformatics/btad591 · DOI
  50. 10.1038/s41467-021-21511-x · DOI · Paper page
  51. 2022, Learning inverse folding from millions of predicted structures.” presented at the Proceedings of the 39th International Conference on Machine Learning, Proceedings of Machine Learning Research, 162, 8946
  52. 10.1126/science.ade2574 · DOI · Paper page
  53. 10.1093/bioinformatics/btac056 · DOI · Paper page

Metadata

Authors
Wang, H., Sun, M., Xie, L., Liu, D., & Zhang, G.
Affiliations
Zhejiang University of Technology (CN)
Corresponding author
Zhang, Guijun
ORCID
Venue
Computational and Structural Biotechnology Journal
Publisher
openRxiv
Published
2024-11-03
Volume
27
Pages
478-487
ISSN
2001-0370 (ISSN-L)
DOI
10.1101/2024.10.31.621211
PMC
PMC11799971
OpenAlex
W4406704559
Cited by
5
References
1
Citations by year
2026: 3 · 2025: 2
FWCI
1.49
Type
journal article
Language
en
Open access
Licence
Locations
  • Computational and Structural Biotechnology Journal (publishedVersion, cc-by-nc-nd)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Natural Science Foundation of China (62173304, 2023R5248)
  • National Key Research and Development Program of China (2022ZD0115103)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Structural Biology › Enzyme Structure and Function
Keywords
protein complex structure prediction, model accuracy assessment, Voronoi tessellation, hydrophobic interactions, graph neural networks, protein language model embeddings, flexible proteins, CASP15

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review