Modeling of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Proteins by Machine Learning and Physics-Based Refinement

Lim Heo; Michael Feig

Protein tertiary structures are essential for understanding their biological mechanisms. Such insight at the molecular level, allows those proteins to be exploited as therapeutic targets by identifying either already approved drug molecules that could be repurposed or discovering new drug candidates via computational methods such as virtual screening. Since the SARS-CoV-2 infection reached pandemic level since early 2020, there is now an urgent need for highresolution structures of this virus. As protein sequences for the virus proteome were determined quickly 1 , some of the protein structures could be obtained experimentally. However, experimental structures of many proteins are still not available to date, leaving prediction via computational methods as the only alternative. SWISS-MODEL 2 could predict tertiary structure models for a subset of proteins by relying on template-based modeling techniques. since many of the genes in the SARS-CoV-2 genome are close homologs to proteins in other organisms with known structures. However, for some of the proteins, template-based modeling is not possible because of lack of experimentally determined close homologs. Recently, the prediction of tertiary structures for proteins where no template structures are available, has been advanced significantly via novel machine learning methods 3 . This approach predicts interresidue distances from multiple sequence alignment via deep learning. Using this approach, DeepMind applying the AlphaFold method 3 to make predictions for six proteins, where close homolog structures are not available 4 . In addition, the Zhang group predicted models for the entire proteome 5 , including targets for which no homologs can be identified, by using the novel C-I-TASSER platform 6 , which is a combined method of contact-based and template-based modeling.