A Deep Integrated Framework for Predicting SARS-CoV2–Human Protein-Protein Interaction

Abstract

Drug repurposing for COVID-19 is now an actively developing area of research. Unavailability of a proper set of interactions between SARS-CoV2 and human host proteins limits the set of possible drug-targets. Toward this, we propose a deep learning based methodology for high confidence interaction prediction between SARS-CoV2 and human host proteins. First, our technique leverage the landmark advantage of Node2Vec to produce a low dimensional embedding from a compiled interaction network that puts SARS-CoV2 proteins, target human host proteins (CoV-host), and the whole human interactome into an encompassing context. Second, we can able to combine information from protein sequence, gene ontology terms and physical interactions information in the prediction task. Third, we proposed a way to rank the host proteins that are potential candidate for target by SARS-CoV2 proteins. Last but not least the method supports a meaningful connection between the predicted proteins and different repurposable drugs to use against COVID-19.