Applications of Deep Learning in Drug Discovery

Abstract

The transformative impact of Deep Learning (DL) on healthcare, particularly in bioinformatics and Drug Discovery, is widely acknowledged. Despite progress, critical gaps persist, including the lack of robust methods for processing large-scale biochemical data, the scarcity of well-annotated datasets, and the limited integration of advanced methodologies with existing repositories. These gaps limit the effectiveness and affordability of conventional drug discovery frameworks, necessitating innovative DL approaches to model these workflows. This thesis emphasizes the adoption of advanced, efficient, and affordable DL approaches to drug discovery, addressing critical healthcare challenges. The primary objective is of leverage DL-based techniques to optimize drug discovery pipelines, reduce costs, and expedite the discovery of novel drug molecules. This work leverages advanced DL algorithms, including Recurrent Neural Networks (RNN), and generative models, such as Variational Autoencoders (VAE), Generative Adversarial Networks (GAN), and Bidirectional Encoder Representations from Transformers (BERT). These models automate key drug discovery tasks, including novel molecule generation, drug-target interaction (DTI) prediction, and molecular property estimation using standard bimolecular datasets. The research findings showed significant improvements in accuracy, precision, efficiency, and cost reduction in drug discovery processes. This research has potential implications for the pharmaceutical industry as it revolutionizes drug discovery through cutting-edge DL-based approaches. These findings contribute to increased drug accessibility and affordability, thereby improving healthcare outcomes in diverse populations.