Predicting Adverse Drug Effects: A Heterogeneous Graph Convolution Network with a Multi-layer Perceptron Approach

Abstract

We exploit a heterogeneous graph convolution network (GCN) combined with a multi-layer perceptron (MLP), which is denoted by GCNMLP, to explore potential side effects of drugs. By inferring the relationship among similar drugs, our approach in silico methods shortens the time consumption in uncovering the side effects unobserved in routine drug prescriptions. In addition, it highlights the relevance in exploring the mechanism of well-documented drugs. Our results predict drug side effects with area under precision-recall (AUPR) curve AUPR = 0.941, substantially outperforming the non-negative matrix factorization (NMF) method with AUPR=0.600. Moreover, new side effects are obtained using the GCNMLP.