Drug Discovery Using Deep Learning Approach

Abstract

Drug development process is a very time consuming and expensive process. Among all the expenses Synthesis and trying out lead analogs being a big contributor to that expenses [Basak, 2012]. Therefore, it is useful to use a computational approach in screening, for lead identification, in order to cover wider chemical space at the same time as decreasing the number of compounds that ought to be synthesized and examined in Vitro testing. The computational approach of compound identification can include a structure based analysis of binding pose and binding energy or predicted biological activity or prediction of drug properties or ligand based screening for drug compounds with similar chemical structure. So ecient prediction of ligand target interaction will speed up the research efforts in drug design. The recent great performance of deep learning in the eld object detection, language translation, speech translation attracted research attention .however deep learning is used as a classier for interaction of drug target pairs and for classification of properties, it is also capable of feature extraction using convolution network and for identification of sequence using recurrent neural networks. Thus, deep learning can be useful in the drug development cycle in prediction of drug properties, de novo drug design and in drug target interaction prediction. Molecular docking is a tool in structure based drug design. The goal of molecular docking is to predict binding anity and pose of binding site. In this study we will focus on ligand target docking. In this paper, we propose, LSTM based network for a drug target prediction system that utilizes a smile string as input features.