Drug-target interaction prediction with hubness-aware machine learning

Abstract

Prediction of interactions between drugs and pharmacological targets is an important task for which various machine learning techniques have been applied recently. Although hubness-aware machine learning techniques are among the most promising recently developed approaches, they have not been used for the prediction of drug-target interactions before. In this paper, we extend the Bipartite Local Model (BLM), one of the most prominent approaches for drug-target interaction prediction. In particular, we propose to use a hubness-aware regression technique, ECkNN, as local model. Furthermore, we propose to represent drugs and targets in the similarity space. In order to assist reproducibility of our work as well as comparison to published results, we perform experiments on widely used publicly available real-world drug-target interaction datasets. The results show that our approach is competitive and, in many cases, superior to state-of-the-art drug-target prediction techniques.