Targeted drug discovery with adversarial graph autoencoders conditioned on gene expression data

• Main Author: Ong, Hiok Hian
• Other Authors: Jagath C Rajapakse
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Science::Medicine::Pharmacy::Pharmaceutical technology
• Online Access: https://hdl.handle.net/10356/166090
• Institution: Nanyang Technological University

Drug discovery has long been an expensive and inefficient process due to the vast chemical compound search space. This process has been iteratively sharpened and refined using computational approaches in order to narrow the scope of search. Separately, machine learning, in particular deep learning, has also made immense progress in many fields such as Computer Vision and Natural Language Processing. In particular, the new wave of generative AI models such as ChatGPT are set to revolutionize many industries in the near future. Recent advancements in deep learning have also achieved much success in many parts of the drug discovery process, from aiding in de novo drug design to modeling quantitative structure activity relationships. This project will focus on applying recent innovations in deep learning based generative AI to allow for controllable generation of potential drug molecules with desired biological properties in order to aid in the drug discovery process. Specifically, gene expression data is introduced into current state of the art de novo drug discovery models and adversarial training is applied to improve it. Additionally, as an effort to standardise the evaluation of the efficacy of future efforts in drug discovery methods catered towards conditional generation of molecules with desirable properties, a set of evaluation methods collated from existing works is proposed alongside a set of active inhibitor molecules for 9 protein targets for benchmarking.