Development of deep learning methods for efficient early stage drug discovery

Abstract

Deep learning techniques attract great attention as a new promising way of reducing time and resources for finding new drug candidates. We address how deep learning techniques can accelerate the process of drug discovery: from hit discovery to lead optimization. In particular, our work focuses on developing deep learning techniques for an accurate prediction of drug-target interaction (DTI) and molecular generative models for designing molecules with desirable molecular properties. As a result, we significantly improved the accuracy of DTI prediction compared to docking and other deep learning techniques. Also, we developed a series of molecular generative models with progressive technical advancement compared to previous models. We believe that our contribution can be the beginning of a long journey toward AI-based drug design.