Drug-target Interaction Prediction Using Machine Learning

Abstract

In the last decade, there has been a major worldwide evolution in the drug discovery and design technologies, in conjunction with the bold emergence of bioinformatics. Different artificial intelligence and machine learning (ML) approaches have been adopted in drug discovery and design, causing a mutation in all their stages, such as drugs and compounds discrimination, compounds classification, and drug-target interaction (DTI) prediction, which is very time consuming. Researchers using positive and negative interactions for DTI prediction usually randomize negative data, considering unknown interaction as negative interactions. This may reduce the accuracy of DTI prediction. In this paper, a new method for drug-target interaction prediction is proposed using improved ML and feature extraction methods for faster and more accurate results. The negative dataset for drugs is generated based on the similarity among drugs, solving the problem of noisy negative data in the literature. The proposed method uses fewer features for drugs and protein targets, which makes DTI prediction and negative data generation faster. The experiments show that the DTIs prediction using the proposed method achieves an average accuracy of 95%, which outperforms the benchmark studies with an average of 3-4 %.