A Network Biology Approach to Predicting Drug Cardiotoxicity

Although modern drugs are designed to regulate the functions of specific target proteins called "drug targets", there can be un-designed "off-target" side effects that make a drug fail to reach the market. Recently, network biology approaches have been to establish a network-pharmacology based drug development paradigm. However, few studies show how to predict adverse drug reactions (ADRs) based on this paradigm. Other approaches to predicting ADRs lack in performances, particularly in prediction specificity. In this study, we present a network biology approach based on support vector machines and logistic regressions to predict drug cardiotoxicity by integrating publicly-available ADR, drug target, and protein-protein interaction (PPI) data. Our approach not only shows better prediction performances (median AUC = 0.771, Accuracy = 0.675, Sensitivity = 0.632, and Specificity = 0.789) with a new in silico model, but also illustrate the significance of incorporating prior knowledge for future ADR assessments.