Machine-Learning Driven Drug Repurposing for COVID-19
The integration of machine learning methods into bioinformatics provides particular benefits in identifying how therapeutics effective in one context might have utility in an unknown clinical context or against a novel pathology. We aim to discover the underlying associations between viral proteins and antiviral therapeutics that are effective against them by employing neural network models. Using the National Center for Biotechnology Information virus protein database and the DrugVirus database, which provides a comprehensive report of broad-spectrum antiviral agents (BSAAs) and viruses they inhibit, we trained ANN models with virus protein sequences as inputs and antiviral agents deemed safe-in-humans as outputs. Model training excluded SARS-CoV-2 proteins and included only Phases II, III, IV and Approved level drugs. Using sequences for SARS-CoV-2 (the coronavirus that causes COVID-19) as inputs to the trained models produces outputs of tentative safe-in-human antiviral candidates for treating COVID-19. Our results suggest multiple drug candidates, some of which complement recent findings from noteworthy clinical studies. Our in-silico approach to drug repurposing has promise in identifying new drug candidates and treatments for other viruses.