International Journal of Scientific Research and Engineering Development

Developing new and effective medications represents one of the most prominent fields in medicine today, of which drug-target interaction prediction represents one of the four major components that characterize drug response. Historically, drug-target interaction prediction has been costly and time-consuming, however, with the plethora of biological and chemical data now available, new strategies such as in silico drug-target interaction prediction, specifically deep learning, have become a critical approach for drug-target interaction predictions. Earlier computational methods were limited by feature engineering, and they lacked the ability to include sophisticated and large biological datasets in their prediction. Deep learning methods have addressed these earlier problems with the ability to auto-construct features and perform on complicated tasks better than previous methods.The review is about ten papers that present a variety of deep learning methods, including hybrid methods, and graph-based methods that showed high accuracy and interpretability. While deep learning-based methods are powerful, many traditional machine learning methods perform adequately with clever feature selection (particularly on smaller datasets). The review also concludes with some critiques of the future challenges in drug-target interaction predictions using deep learning, including challenges with hybrid methods, approaches to mitigating imbalance datasets, and interpretability to enhance rational drug design.