Drug-Drug Adverse Reactions Prediction Based On Signed Network

Abstract

When used to treat patients' diseases, drugs may harm their health. The more we know about drug-drug adverse drug reactions (DDADRs), the better we can avoid accidents. As there are thousands of drugs on the pharmaceutical market, it is impossible to perform experiments in the laboratory to detect the adverse effects caused by the drug-drug interactions (DDIs). Therefore, data-driven methods have become popular. Although there are many deep neural networks (DNN) based models for predicting adverse drug reactions (ADRs), they all described the drug-drug relationships with unsigned networks which ignore the polarity of the drug-drug interactions. Therefore, this paper proposes a model GS-ADR which not only considers the relationship between a variety of drugs, but also depicts the polarity of the interactions between drugs. We find that when the positive and negative relationship between drugs considered at the same time, the feature representation of the drugs is more effective, which is helpful for predicting the drug-drug relations. Experimental results show that our proposed method achieves the state-of-the-art results compared with several deep learning models based link prediction models.