Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Deep-learning-enabled protein–protein interaction analysis for prediction of SARS-CoV-2 infectivity and variant evolution

Nature Medicine volume 29, pages 2007–2018 (2023)Cite this article

Subjects

Abstract

Host–pathogen interactions and pathogen evolution are underpinned by protein–protein interactions between viral and host proteins. An understanding of how viral variants affect protein–protein binding is important for predicting viral–host interactions, such as the emergence of new pathogenic SARS-CoV-2 variants. Here we propose an artificial intelligence-based framework called UniBind, in which proteins are represented as a graph at the residue and atom levels. UniBind integrates protein three-dimensional structure and binding affinity and is capable of multi-task learning for heterogeneous biological data integration. In systematic tests on benchmark datasets and further experimental validation, UniBind effectively and scalably predicted the effects of SARS-CoV-2 spike protein variants on their binding affinities to the human ACE2 receptor, as well as to SARS-CoV-2 neutralizing monoclonal antibodies. Furthermore, in a cross-species analysis, UniBind could be applied to predict host susceptibility to SARS-CoV-2 variants and to predict future viral variant evolutionary trends. This in silico approach has the potential to serve as an early warning system for problematic emerging SARS-CoV-2 variants, as well as to facilitate research on protein–protein interactions in general.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on Springer Link
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Protein–protein binding affinities and variant evolution prediction with UniBind.
Fig. 2: UniBind performance on protein complex affinity prediction.
Fig. 3: UniBind performance for predicting the impact of SARS-CoV-2 RBD mutations on ACE2 and RBD antibody affinity.
Fig. 4: UniBind performance for predicting the binding affinity between the S-protein and ACE2 variants.
Fig. 5: Characteristics and forecast of SARS-CoV-2 evolution.

Data availability

All input datasets are freely available from public sources.

Code availability

The deep-learning models were developed and deployed using standard model libraries and the PyTorch framework. Custom codes were specific to our development environment and used primarily for data input/output and parallelization across computers and graphics processors. Code is available at GitHub (https://github.com/UniBind/UniBind).

References

  1. Menachery, V. D. et al. A SARS-like cluster of circulating bat coronaviruses shows potential for human emergence. Nat. Med. 21, 1508–1513 (2015).

    CAS  PubMed  PubMed Central  Google Scholar 

  2. Starr, T. N. et al. ACE2 binding is an ancestral and evolvable trait of sarbecoviruses. Nature 603, 913–918 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  3. Walls, A. C. et al. Structure, function, and antigenicity of the SARS-CoV-2 spike glycoprotein. Cell 181, 281–292.e286 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  4. Korber, B. et al. Tracking changes in SARS-CoV-2 spike: evidence that D614G increases infectivity of the COVID-19 virus. Cell 182, 812–827.e819 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  5. Thomson, E. C. et al. Circulating SARS-CoV-2 spike N439K variants maintain fitness while evading antibody-mediated immunity. Cell 184, 1171–1187.e1120 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  6. Hill, V. et al. The origins and molecular evolution of SARS-CoV-2 lineage B.1.1.7 in the UK. Virus Evol. 8, veac080 (2022).

  7. Mlcochova, P. et al. SARS-CoV-2 B.1.617.2 Delta variant replication and immune evasion. Nature 599, 114–119 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  8. Viana, R. et al. Rapid epidemic expansion of the SARS-CoV-2 Omicron variant in southern Africa. Nature 603, 679–686 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  9. Martin, D. P. et al. Selection analysis identifies clusters of unusual mutational changes in Omicron lineage BA.1 that likely impact spike function. Mol. Biol. Evol. 39, msac061 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  10. Wang, Q. et al. Alarming antibody evasion properties of rising SARS-CoV-2 BQ and XBB subvariants. Cell https://doi.org/S0092867422015318 (2022).

  11. Yue, C. et al. ACE2 binding and antibody evasion in enhanced transmissibility of XBB.1.5. Lancet Infect. Dis. 23, 278–280 (2023).

  12. Mannar, D. et al. SARS-CoV-2 Omicron variant: antibody evasion and cryo-EM structure of spike protein-ACE2 complex. Science 375, 760–764 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  13. Cao, Y. et al. Omicron escapes the majority of existing SARS-CoV-2 neutralizing antibodies. Nature 602, 657–663 (2022).

    CAS  PubMed  Google Scholar 

  14. Jankauskaitė, J., Jiménez-García, B., Dapkūnas, J., Fernández-Recio, J. & Moal, I. H. SKEMPI 2.0: an updated benchmark of changes in protein–protein binding energy, kinetics and thermodynamics upon mutation. Bioinformatics 35, 462–469 (2019).

    PubMed  Google Scholar 

  15. Taft, J. M. et al. Deep mutational learning predicts ACE2 binding and antibody escape to combinatorial mutations in the SARS-CoV-2 receptor-binding domain. Cell 185, 4008–4022.e4014 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  16. Wang, B. & Gamazon, E. R. Modeling mutational effects on biochemical phenotypes using convolutional neural networks: application to SARS-CoV-2. iScience 25, 104500 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  17. Elbe, S. & Buckland-Merrett, G. Data, disease and diplomacy: GISAID’s innovative contribution to global health. Glob. Chall. 1, 33–46 (2017).

    PubMed  PubMed Central  Google Scholar 

  18. Hie, B., Zhong, E. D., Berger, B. & Bryson, B. Learning the language of viral evolution and escape. Science 371, 284–288 (2021).

    CAS  PubMed  Google Scholar 

  19. Maher, M. C. et al. Predicting the mutational drivers of future SARS-CoV-2 variants of concern. Sci. Transl. Med. 14, eabk3445 (2022).

    CAS  PubMed  Google Scholar 

  20. Obermeyer, F. et al. Analysis of 6.4 million SARS-CoV-2 genomes identifies mutations associated with fitness. Science 376, 1327–1332 (2022).

  21. Rodriguez-Rivas, J., Croce, G., Muscat, M. & Weigt, M. Epistatic models predict mutable sites in SARS-CoV-2 proteins and epitopes. Proc. Natl Acad. Sci. USA 119, e2113118119 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  22. Beguir, K. et al. Early computational detection of potential high risk SARS-CoV-2 variants. Comput. Biol. Med. 155, 106618 (2023).

  23. Chan, K. K. et al. Engineering human ACE2 to optimize binding to the spike protein of SARS coronavirus 2. Science 369, 1261–1265 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  24. Starr, T. N. et al. Shifting mutational constraints in the SARS-CoV-2 receptor-binding domain during viral evolution. Science 377, 420–424 (2022).

    CAS  PubMed  Google Scholar 

  25. Starr, T. N. et al. Deep mutational scanning of SARS-CoV-2 receptor binding domain reveals constraints on folding and ACE2 binding. Cell 182, 1295–1310.e1220 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  26. Esposito, D. et al. MaveDB: an open-source platform to distribute and interpret data from multiplexed assays of variant effect. Genome Biol. 20, 223 (2019).

    PubMed  PubMed Central  Google Scholar 

  27. Han, P. et al. Receptor binding and complex structures of human ACE2 to spike RBD from omicron and delta SARS-CoV-2. Cell 185, 630–640.e610 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  28. Han, P. et al. Molecular insights into receptor binding of recent emerging SARS-CoV-2 variants. Nat. Commun. 12, 6103 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  29. Higuchi, Y. et al. Engineered ACE2 receptor therapy overcomes mutational escape of SARS-CoV-2. Nat. Commun. 12, 3802 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  30. Liu, L. et al. Striking antibody evasion manifested by the Omicron variant of SARS-CoV-2. Nature 602, 676–681 (2022).

    CAS  PubMed  Google Scholar 

  31. Wang, R. et al. Analysis of SARS-CoV-2 variant mutations reveals neutralization escape mechanisms and the ability to use ACE2 receptors from additional species. Immunity 54, 1611–1621.e1615 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  32. Rodrigues, C. H. M., Myung, Y., Pires, D. E. V. & Ascher, D. B. mCSM-PPI2: predicting the effects of mutations on protein-protein interactions. Nucleic Acids Res. 47, W338–W344 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

  33. Pires, D. E. V. & Ascher, D. B. mCSM-AB: a web server for predicting antibody-antigen affinity changes upon mutation with graph-based signatures. Nucleic Acids Res. 44, W469–W473 (2016).

    CAS  PubMed  PubMed Central  Google Scholar 

  34. Xiong, P., Zhang, C., Zheng, W. & Zhang, Y. BindProfX: assessing mutation-induced binding affinity change by protein interface profiles with pseudo-counts. J. Mol. Biol. 429, 426–434 (2017).

    CAS  PubMed  Google Scholar 

  35. Yang, B., Li, K., Zhong, X. & Zou, J. Implementation of deep learning in drug design. MedComm Future Med. 1, e18 (2022).

  36. Bhattacharjee, M. J. et al. Identifying primate ACE2 variants that confer resistance to SARS-CoV-2. Mol. Biol. Evol. 38, 2715–2731 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  37. Ye, F. et al. S19W, T27W, and N330Y mutations in ACE2 enhance SARS-CoV-2 S-RBD binding toward both wild-type and antibody-resistant viruses and its molecular basis. Signal Transduct. Target. Ther. 6, 343 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  38. Damas, J. et al. Broad host range of SARS-CoV-2 predicted by comparative and structural analysis of ACE2 in vertebrates. Proc. Natl Acad. Sci. USA 117, 22311–22322 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  39. Shi, J. et al. Susceptibility of ferrets, cats, dogs, and other domesticated animals to SARS-coronavirus 2. Science 368, 1016–1020 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  40. Oude Munnink, B. B. et al. Transmission of SARS-CoV-2 on mink farms between humans and mink and back to humans. Science 371, 172–177 (2021).

    CAS  PubMed  Google Scholar 

  41. Cameroni, E. et al. Broadly neutralizing antibodies overcome SARS-CoV-2 Omicron antigenic shift. Nature 602, 664–670 (2022).

    CAS  PubMed  Google Scholar 

  42. Hong, Q. et al. Molecular basis of receptor binding and antibody neutralization of Omicron. Nature 604, 546–552 (2022).

    CAS  PubMed  Google Scholar 

  43. Meng, B. et al. Altered TMPRSS2 usage by SARS-CoV-2 Omicron impacts infectivity and fusogenicity. Nature 603, 706–714 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  44. Triveri, A. et al. SARS-CoV-2 spike protein mutations and escape from antibodies: a computational model of epitope loss in variants of concern. J. Chem. Inf. Model. 61, 4687–4700 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  45. Gruell, H. et al. Neutralisation sensitivity of the SARS-CoV-2 omicron BA.2.75 sublineage. Lancet Infect. Dis. 22, 1422–1423 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  46. Wu, L. et al. SARS-CoV-2 Omicron RBD shows weaker binding affinity than the currently dominant Delta variant to human ACE2. Signal Transduct. Target. Ther. 7, 8 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  47. Imai, M. et al. Efficacy of antiviral agents against Omicron subvariants BQ.1.1 and XBB. N. Engl. J. Med. 388, 89–91 (2023).

  48. Cao, Y. et al. BA.2.12.1, BA.4 and BA.5 escape antibodies elicited by Omicron infection. Nature 608, 593–602 (2022).

  49. Callaway, E. How months-long COVID infections could seed dangerous new variants. Nature 606, 452–455 (2022).

    CAS  PubMed  Google Scholar 

  50. Sonnleitner, S. T. et al. Cumulative SARS-CoV-2 mutations and corresponding changes in immunity in an immunocompromised patient indicate viral evolution within the host. Nat. Commun. 13, 2560 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  51. Rives, A. et al. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences. Proc. Natl Acad. Sci. USA 118, e2016239118 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  52. Lyngse, F. P. et al. Household transmission of SARS-CoV-2 Omicron variant of concern subvariants BA.1 and BA.2 in Denmark. Nat. Commun. 13, 5760 (2022).

  53. Jian, F. et al. Further humoral immunity evasion of emerging SARS-CoV-2 BA.4 and BA.5 subvariants. Lancet Infect. Dis. 22, 1535–1537 (2022).

  54. Yamasoba, D. et al. Virological characteristics of the SARS-CoV-2 Omicron BA.2 spike. Cell 185, 2103–2115.e19 (2022).

  55. Cheng, S. M. S. et al. Neutralizing antibodies against the SARS-CoV-2 Omicron variant BA.1 following homologous and heterologous CoronaVac or BNT162b2 vaccination. Nat. Med. 28, 486–489 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  56. Iketani, S. et al. Antibody evasion properties of SARS-CoV-2 Omicron sublineages. Nature 604, 553–556 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  57. Yu, J. et al. Neutralization of the SARS-CoV-2 Omicron BA.1 and BA.2 variants. N. Engl. J. Med. 386, 1579–1580 (2022).

    PubMed  Google Scholar 

  58. Andrews, N. et al. Covid-19 vaccine effectiveness against the Omicron (B.1.1.529) variant. N. Engl. J. Med. 386, 1532–1546 (2022).

    CAS  PubMed  Google Scholar 

  59. Cao, Y. et al. Imprinted SARS-CoV-2 humoral immunity induces convergent Omicron RBD evolution. Nature 614, 521–529 (2023).

    CAS  PubMed  Google Scholar 

  60. Hadfield, J. et al. Nextstrain: real-time tracking of pathogen evolution. Bioinformatics 34, 4121–4123 (2018).

    CAS  PubMed  PubMed Central  Google Scholar 

  61. Sarfati, H., Naftaly, S., Papo, N. & Keasar, C. Predicting mutant outcome by combining deep mutational scanning and machine learning. Proteins 90, 45–57 (2022).

    CAS  PubMed  Google Scholar 

  62. Huang, X., Pearce, R. & Zhang, Y. EvoEF2: accurate and fast energy function for computational protein design. Bioinformatics 36, 1135–1142 (2020).

    CAS  PubMed  Google Scholar 

  63. Xia, K. & Wang, J. Recent advances of transformers in medical image analysis: a comprehensive review. MedComm Future Med. 2, e38 (2023).

  64. Gao, Y., Zhan, J. & Yu, A. C. H., Yu. Understanding by design: implementing deep learning from protein structure prediction to protein design. MedComm Future Med. 1, e22 (2022).

  65. Barlow, K. A. et al. Flex ddG: Rosetta Ensemble-based estimation of changes in protein-protein binding affinity upon mutation. J. Phys. Chem. B 122, 5389–5399 (2018).

    CAS  PubMed  PubMed Central  Google Scholar 

  66. Wehenkel, A. & Louppe, G. Unconstrained monotonic neural networks. Preprint at https://arxiv.org/abs/1908.05164 (2021).

  67. Shan, S. et al. Deep learning guided optimization of human antibody against SARS-CoV-2 variants with broad neutralization. Proc. Natl Acad. Sci. USA 119, e2122954119 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

  68. Ulrich, L. et al. Enhanced fitness of SARS-CoV-2 variant of concern Alpha but not Beta. Nature 602, 307–313 (2022).

    CAS  PubMed  Google Scholar 

Download references

Acknowledgements

This study was funded by the National Natural Science Foundation of China (grant no. 62272055), the New Cornerstone Science Foundation through the XPLORER PRIZE, Young Elite Scientists Sponsorship Program by CAST (grant no. 2021QNRC001), the Major Key Project of PCL (grant no. PCL2021A15), Guangzhou National Laboratory, Macau University of Science and Technology, the Macau Antibody Protection Study (MAPS) and the Macau Science and Technology Development Fund (grant nos. 0007/2020/AFJ, 0070/2020/A2, 0109/2020/A3 and 0003/2021/AKP).

Author information

Author notes

  1. These authors contributed equally: Guangyu Wang, Xiaohong Liu, Kai Wang, Yuanxu Gao, Gen Li, Daniel T. Baptista-Hon.

Authors and Affiliations

  1. State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China

    Guangyu Wang, Zeyu Jiang & Ping Zhang

  2. Instutite for Artificial Intelligence in Medicine and Faculty of Medicine, Macau University of Science and Technology, Macau, China

    Xiaohong Liu, Daniel T. Baptista-Hon, Xiaohong Helena Yang, Wa Hou Tai, Linling Cheng, Manson Fok, Ligong Lu & Kang Zhang

  3. UCL Cancer Institute, University College London, London, UK

    Xiaohong Liu

  4. Department of Big Data and Biomedical Artificial Intelligence, National Biomedical Imaging Center, College of Future Technology, Peking University and Peking-Tsinghua Center for Life Sciences, Beijing, China

    Kai Wang & Kang Zhang

  5. Guangzhou National Laboratory, Guangzhou, China

    Yuanxu Gao, Gen Li & Kang Zhang

  6. Guangzhou Women and Children’s Medical Center, Guangzhou, China

    Gen Li

  7. Zhuhai International Eye Center and Provincial Key Laboratory of Tumor Interventional Diagnosis and Treatment, Zhuhai People’s Hospital and the First Affiliated Hospital of Faculty of Medicine, Macau University of Science and Technology, Guangdong, China

    Daniel T. Baptista-Hon, Linling Cheng, Ligong Lu & Kang Zhang

  8. Nuffield Laboratory of Ophthalmology, Department of Clinical Neurosciences, University of Oxford, Oxford, UK

    Kanmin Xue

  9. Departments of Biology and Chinese Medicine, Hong Kong Baptist University, Hong Kong SAR, China

    Johnson Yiu-Nam Lau

  10. State Key Laboratory of Biotherapy, West China Hospital, Sichuan University, Chengdu, China

    Shengyong Yang

Authors
  1. Guangyu Wang
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Xiaohong Liu
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Kai Wang
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Yuanxu Gao
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Gen Li
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Daniel T. Baptista-Hon
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Xiaohong Helena Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Kanmin Xue
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Wa Hou Tai
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Zeyu Jiang
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Linling Cheng
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Manson Fok
    View author publications

    You can also search for this author in PubMed Google Scholar

  13. Johnson Yiu-Nam Lau
    View author publications

    You can also search for this author in PubMed Google Scholar

  14. Shengyong Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  15. Ligong Lu
    View author publications

    You can also search for this author in PubMed Google Scholar

  16. Ping Zhang
    View author publications

    You can also search for this author in PubMed Google Scholar

  17. Kang Zhang
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

X.L., K.W., Y.G., G.L., D.T.B.-H., X.H.Y., K.X., W.H.T., Z.J., L.C., M.F., J.Y.-N.L., S.Y., L.L., P.Z., G.W. and K.Z. collected and/or analyzed the data. K.Z. and G.W. conceived and supervised the project. K.Z., X.L., J.Y.-N.L., Y.G., X.H.Y. and G.W. wrote and/or revised the paper. All authors discussed the results and reviewed the paper.

Corresponding authors

Correspondence to Guangyu Wang or Kang Zhang.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Medicine thanks Eric Gamazon and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Michael Basson, in collaboration with the Nature Medicine team.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Architectural details of the geometry and energy attention module.

Arrows show the information flow. Vector names: XS: input sequence feature, XE: input energy feature, XG: input geometric feature, XN: neighbor sequence feature, TN: neighbor translation, T: local transition, R: neighbor translation, \({X}_{S}^{{\prime} }\): output sequence feature. L: length of the amino acid sequence, LN: No. of nearest neighbor residues in the graph, Nh: No. of heads in the multi-head attention. Dimension names: dS: sequence feature, dE: energy feature, dG: geometric feature, dQ: query vector, dK: key vector, dV: value vector. Operators: ⊙: element-wise multiplication, ⊕: element-wise addition. Functions: k−NN: search for k-nearest neighbors, \(\text{Linear}\) \({d}_{1}\to {d}_{2}\): fully connected layer with an input dimension of d1 and an output dimension of d2, \(f\left(x\right):3\to {d}_{G}\): geometric feature function, specifically, \(f\left(x\right)={concat}(x,{\vec{n}}_{x},{||x|}{|}_{2})\) in BindFormer, where \({\vec{n}}_{x}\) is the normal vector of x.

Extended Data Fig. 2 Validation of the AI’s performance on protein complex affinity prediction.

a-b, mean absolute error of between calculated and experimental values of changes in binding affinity in SKEMPI V2.0, grouped by a, the number of mutations (1 to 7+ mutations, n = 4069, 796, 287, 218, 116, 59 and 184, respectively; error bar denotes standard deviation); and b, by structure’s resolution. c-d, Regression analysis between predicted scores and experimental scores in MaveDB. c, Performance on cohesin-dockerin binding score in Clostridium cellulolyticum, d, Performance on PSD95 protein binding score with small peptide ligand. e, Performance comparison in MaveDB. f, Regression performance between predicted scores and experimental scores on IGBPG (immunoglobulin G-binding β1 domain of streptococcal protein G) dataset. Error bar, standard deviation.

Extended Data Fig. 3 UniBind performance on measuring effects on ACE2 binding of mutations to SARS-CoV-2 RBD.

a-d, Regression performance of affinity prediction of RBD mutation effects on different SARS-CoV-2 variants including a, Alpha (N501Y), b, Beta (K417N+E484K+N501Y), c, Delta (L452R+T478K), d, Eta (E484K). MAE, mean absolute error; R2, coefficient of determination; PCC, Pearson’s correlation coefficient.

Extended Data Fig. 4 Prediction of effects on antibody binding of mutations to SARS-CoV-2 RBD.

a–d, Stratified analysis of regression performance on 4 classes of neutralization antibodies which were grouped according to Cao et al. e, f, Heatmap of experimental (e) and predicted (f) escape score matrix upon mutations of RBD to different antibodies. Brightness represents the escape score. A brighter dot indicates that the mutation on site position of x-axis is more likely to lead to higher immune escape for antibody of y-axis. MAE, mean absolute error; R2, coefficient of determination; PCC, Pearson’s correlation coefficient.

Extended Data Fig. 5 Prediction of antibody binding of Variant-of-Concerns (VOCs).

a–h, predicted escape scores of antibodies for each VOC are shown in the boxplot. For each analysis, antibodies were separated into two groups that can be escaped or not escaped by SARS-CoV-2 variants according to relative literature. The Center line indicates median; box limits indicate upper and lower quartiles; whiskers indicate 1.5x interquartile range; points indicate outliers; P values less than 0.05, 0.01, 0.001, 0.0001 are summarized with one to four asterisks, respectively. The number of non-escape and escape variants for each variant (Alpha: 18, 1 (a); Delta: 12, 1 (b); Epsilon: 10, 1 (c); lota: 8, 2 (d); Beta: 11, 8 (e); Gamma: 12, 7 (f); Omicron_BA1: 2, 11 (g); Omicron_R346K: 2, 9 (h); i, ROC curves of neutralization escape prediction.

Extended Data Fig. 6 UniBind performance on predicting the binding affinity between the SARS-CoV-2 and ACE2 mutations (hACE2 and cross-species).

a and b, Magnified views of AI-predicted 3D structures of wild type/mutant ACE2 in complex with the wildtype SARS-CoV-2. a, ACE2 mutant carrying residue N330Y interfaces with corresponding residues P499, T500 on SARS-CoV-2 RBD. b, ACE2 mutant carrying residue Q42/L42 interfaces with related residues Q498, Y449 on SARS-CoV-2 RBD.

Extended Data Fig. 7 Validation and prediction of variant function.

a, Correlation analysis between reported fitness and affinity-based evolutionary score (evo-score). b, Validation of immune escape prediction on the Omicron sublineage. Yellow dots (left y axis) indicate log transformed fifty-percent inhibitory dilutions (ID50s) of pseudovirus neutralization assay (n=30), curated from Gruell et al. Blue dots (right y axis) indicated log transformed escape scores of four variants against by monoclonal antibodies. Columns shows mean of the data. Error bar shows standard deviation. Differences between variants were tested by two-tailed Student’s t-test. c, AI’s Quantification of the variant’s function during evolutionary process in one COVID-19 patient. Lower panel: variants detected in a COVID-19 patient from 73rd days to 207th day after infected. The variants classified into ‘Found in VOC’ (variant of concern) and ‘Not found in VOC’, denoted with ‘o’ and ‘+’ respectively. Upper panel: The curves of the quantified variant’s function predicted by our UniBind, including the ACE2 affinity, antibody escape, and evo-score, colored with blue, orange, and green, respectively. d, e, Change in affinity and antibody escape scores of predicted SARS-CoV-2 mutations, ranked according to the magnitude of change. Mutations which are found in the immunocompromised patient, for example E484K and E484Q ranked high on our antibody escape prediction (top 0.1% and 0.8%, respectively). In addition, N501Y, which is also found in the immunocompromised patient, was ranked second in our predictions (top 0.1%).

Extended Data Fig. 8 Validation and prediction of variant fitness.

a-c, Heat maps generated by UniBind deep mutational scanning on S-ACE2 binding affinity values (a), antibody escape scores (b), and evo-score values (c). d, A diagram depicting the variants and their mutation load in the sub-lineage of BQ.1.1 (adapted from Nextstrain70). e, f, The top 50 predicted mutations ranked by an immune escape score. The AI model correctly predicted key mutation S494P (orange) which is present in many VOCs such as BQ.1.1.11, BQ.1.1.12, BQ.1.1.13, BQ.1.1.34, DT.1. (created based on nextstrain.org).

Extended Data Table 1 Comparison of various methods on prediction performance in the SKEMPI 2.0 set with mutation-level validation
Full size table
Extended Data Table 2 Comparison of various methods on prediction performance in the SKEMPI 2.0 set with complex-level validation
Full size table

Supplementary information

Supplementary Information

Supplementary Tables 1 and 2.

Reporting Summary

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Wang, G., Liu, X., Wang, K. et al. Deep-learning-enabled protein–protein interaction analysis for prediction of SARS-CoV-2 infectivity and variant evolution. Nat Med 29, 2007–2018 (2023). https://doi.org/10.1038/s41591-023-02483-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41591-023-02483-5

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing