A DEEP LEARNING APPROACH FOR DRUG-TARGET AFFINITY PREDICTIONCOMPREHENSIVE TRANSACTIVE ENERGY MANAGEMENT SYSTEMS

Abstract

The identification of drug-target interaction (DTI) is a crucial part of the drug discovery and development process. In vitro and in vivo experiments for drug target validation and screening are, however, very expensive and take a lot of time to complete. There experiment on large scale are unfeasible, thus there is a huge demand for the development of computational in silico alternatives for DTI prediction. Several statistical and machine learning-based methods have been developed over time that focused on the binary classification of DTI. However, these interactions are very complex, as there is a dynamic fluctuation present between the protein and the bound compound and a continuous mutually flexible adjustment, which needs to be simplified by reaching an equilibrium state characterised by well established binding affinity descriptor. The exact estimation of the binding affinity in the DTI still remains a challenge to this day. Various machine and deep learning methodologies have been developed that utilize different feature representation approaches for both compounds and proteins. These algorithms generally utilize as input limited chemical information, which may not be meaningful and intuitive enough to be used as an effective descriptor. In this work I am addressing the limitation of current methods by introducing a deep learning-based model that makes use of chemical representations of the molecules. Results of experiments on two benchmark datasets demonstrate that the proposed model outperforms the baseline model, which is one of the state-of-the-art methods in the drug-target affinity (DTA) prediction field.