Applications of Deep Learning in Drug Discovery

Abstract

Armed with advances in computational resources and high data throughput, artificial intelligence techniques have achieved remarkable success in diverse application areas in past decade. In recent years the field of pharmaceutical drug discovery has seen upsurge of deep learning applications that go beyond bioactivity predictive models and aid in various facets of drug discovery process. One of the biggest strengths of deep neural networks is their ability to learn from complex nonlinear data without explicit need for handpicking the features. This chapter aims to provide an overview of deep learning methods and their applications in the drug design field. The chapter begins by introducing concepts of machine learning, artificial neural network, and deep learning. Advances in deep neural architecture are discussed with examples of convolutional neural networks (CNNs), recurrent neural networks (RNNs), variational autoencoders (VAEs), and generative adversarial networks (GANs). Application examples of these architectures such as RNN-based variational autoencoders for de novo molecular design, natural language processing, use of adversarial network in GANs for obtaining valid molecular designs, bioactivity prediction, and image-based profiling of bioassays using CNNs are reviewed to bring out variety of drug design challenges being addressed using deep learning techniques.