Repurposing Drugs for COVID-19 by Graph Convolutional Network
Rongting Yue, Abhishek Dutta · Research Square · 2023
Abstract Computational approaches for drug repurposing speed up the process of finding drugs for fast-spreading diseases such as COVID-19, which is challenging due to unknown disease mechanisms and pathogens. Here we present a graph learning-based pipeline to identify the potential drugs that are likely to intervene in COVID-19 virus infection using RNA-seq data and molecular interactions and features. We do so by detecting disease-related genes and pathways first, then constructing a heterogeneous graph model that integrates protein-protein interactions, drug-protein interaction and drug-target interactions, and finally applying graph convolutional network on the graph model to predict potential drug molecules. With this method, we predict 28 drugs that have the potential to bind with disease-related proteins. Molecular docking shows that 2 drugs, Flunisolide and Adapalene, have a high possibility to be used for COVID-19 by stably binding with the products of disease-related genes STAT1 and AKT1, respectively, which will be later confirmed by experiments. The proposed method applies to drug repurposing for any diseases with gene expression data, and it is expected to expand our understanding of novel disease mechanisms and select drug candidates for therapy design.