The Covid-19 pandemic revealed the need for a broad-spectrum antiviral medication for the human coronavirus disease. We utilized two computational engines to predict repurposing drug candidates and validated those predictions experimentally in a cell-based model. First, we applied graph convolutional networks (GCN) trained on multiscale host-virus interactome data to identify novel human targets relevant for COVID-19. Then, we queried our drug-repurposing database, PolypharmDB, with those targets, in order to identify novel off-targets interactions of approved and investigational therapeutics. PolypharmDB is a precompiled database of interaction predictions between 8525 human proteins and 10,244 small molecule compounds with prior clinical evaluation. The predictions are evaluated by using MatchMaker™, our deep learning model that predicts drug-target interactions, trained using paired structural features of the drug with the 3D structural features of the protein binding sites. Next, we tested the predicted 26 candidates for antiviral activity against 229E human alphacoronavirus in a cell-based experimental model. We found that four clinically-relevant repurposing drug candidates predicted by the in silico analyses significantly attenuated 229E infection. In particular, the MET inhibitor capmatinib had a potent and broad antiviral activity against several coronaviruses in a MET-independent manner. We have also proposed novel roles for host proteins such as IRAK1/4 in supporting coronavirus infection, which can inform further drug discovery studies.

An application of an AI-based drug-repurposing platform, PolypharmDB, to uncover antiviral activity of an approved drug

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.