Local Imbalance based Ensemble for Predicting Interactions between Novel Drugs and Targets

B. Liu, K. Pliakos, C. Vens, G. Tsoumakas. (2020) Local Imbalance based Ensemble for Predicting Interactions between Novel Drugs and Targets. Machine Learning for Pharma and Healthcare Applications Workshop (PharML 2020)

Computational prediction of drug-target interactions (DTI) reduces the number of candidate drugs to be verified by the tedious and costly experimental approach and expedites the drug discovery process. The most challenging task for computational DTI prediction methods is to predict interactions between new drugs and new targets due to the unavailability of interacting information for both new drugs and new targets. Although there are several methods that could predict interactions in new drug-target pairs, the accuracy of their predicting results is not adequate. To improve the performance of existing approaches, we propose three ensemble DTI prediction strategies that could accompany any DTI prediction method. The proposed ensemble approaches consist of several DTI prediction models learned on training subsets which have been defined by different sampling strategies. Experiments were conducted on four benchmark datasets and the obtained results indicate that the local imbalance-aware sampling strategy is the most effective.