Abstract
Adverse drug reaction (ADR) is defined as the unwanted and harmful reaction to a drug. ADR can significantly increase the clinical and economic burdens on the public. Additionally, an increased usage in polypharmacy was observed that makes ADR problems more significant. Thus, monitoring and prediction of ADRs becomes necessary. The purpose of this study is to predict ADRs not only for single drug but also for polypharmacy. A computational network-based approach, a random walk model, was proposed and applied on integrated chemical and phenotypic networks which were built from four types of data. Evaluations for single drug ADR prediction is based on the 5-fold cross validation. Different combinations of four types of data were evaluated. The results suggested that by combining four types of data appropriately, our ADR prediction model for single drug achieved the performance of 0.9431 for accuracy and 0.5111 for F1-score. For polypharmacy experiment, our evaluations were based on considering ranks of ADRs. The results suggested that our method had potential ability to predict ADRs for polypharmacy.