RUSDR: Class Imbalance-aware Ensemble Learning for Drug Repurposing

A common problem in many application domains such as drug repurposing, i.e. identifying new indications for known drugs, is class imbalance. There are some approaches to dealing with class imbalance problem, including sampling methods, cost-sensitive learning methods, one-class classification methods, and ensemble learning methods. In this paper, we proposed an approach, called RUSBoosted Tree Drug Repurposing (RUSDR), to predict candidate drug-indications. Initially, a multi-view data of drugs and diseases were used for feature vector construction. Then, we used an ensemble learning-based algorithm capable of addressing class imbalance problem. The experimental results showed improvements in prediction performance over the existing works and the importance of dealing with the class imbalance problem in the data.