MGATRx: Discovering Drug Repositioning Candidates Using Multi-view Graph Attention.

In-silico drug repositioning or predicting new indications for approved or late-stage clinical trial drugs is a resourceful and time-ecient strategy in drug discovery. However, inferring novel candidate drugs for a disease is challenging, given the heterogeneity and sparseness of the underlying biological entities and their relationships (e.g., disease/drug annotations). By integrating drug-centric and disease-centric annotations as multiviews, we propose a multi-view graph attention network for indication discovery (MGATRx). Unlike most current similaritybased methods, we employ graph attention network on the heterogeneous drug and disease data to learn the representation of nodes and identify associations. MGATRx outperformed four other state-of-art methods used for computational drug repositioning. Further, several of our predicted novel indications are either currently investigated or are supported by literature evidence, demonstrating the overall translational utility of MGATRx.