Predicting Drug-drug Interactions Using Heterogeneous Graph Attention Networks

Abstract

Drug-Drug Interactions (DDIs) can alter a drug's efficacy and lead to adverse effects. Predicting potential DDIs during clinical trials is challenging; thus, computational methods are gaining prominence. We present a novel DDI prediction method, constructing a Heterogeneous Information Network (HIN) integrating biomedical entities such as drugs, proteins, and side effects. Our end-to-end model, HAN-DDI, based on a heterogeneous graph attention network, demonstrates superior accuracy in predicting DDIs, surpassing existing methods.