An Improved Approach for Predicting Drug Target Interactions

Abstract

The identification of drug protein associations assists the exploration of novel drugs, drug repurposing and drug side effect identification. The experimental evaluation of these interactions requires extensive capital and money. Thus, in-silico computational methods are being developed to aid the interaction prediction. These techniques have been broadly grouped into similarity based approaches and feature based approaches. This paper proposes a novel feature based approach to identify the probable drug protein communications. The method is based on Support Vector Machine classifier. Support Vector Machines have shown a satisfactory performance in many applications related to the pharmacology domain. To further improve the accuracy and reduce the computational complexity, dimensionality reduction by PCA has been proposed. The proposed technique achieves an AUC score of 0.822. The method has been compared to various other state of the art methods based on their respective AUC scores. The comparison has shown that the proposed approach has a better performance in contrast to the other techniques.