The School of Computer Science is pleased to present…
Integrating Explainable AI Techniques with Generative Adversarial Networks for Enhanced De Novo Drug Generation
MSc Thesis Proposal by: Harshitha Kosuru
Date: Wednesday, 14 August 2024
Time: 1:00 pm
Location: Essex Hall 122
Generative AI models, such as Generative Adversarial Networks (GANs), are increasingly becoming significant in drug discovery, enabling the generation of novel molecular structures. Despite their potential, these models face critical issues like transparency and mode collapse. The opaque nature of neural networks undermines trust among researchers and complicates the optimization of the process and results. Additionally, mode collapse results in the generation of limited molecular varieties, failing to explore the full diversity of the chemical space. These issues hinder model optimization, impede compliance with stringent regulatory standards, and lead to inefficiencies in drug development. Tackling these challenges is crucial to improving the reliability and effectiveness of AI-driven drug discovery. This research integrates Explainable Artificial Intelligence (XAI) techniques within GAN-based Molecule Generative Model frameworks. The objective is to enhance transparency, mitigate mode collapse, improve optimization, and provide clearer insights into the AI-driven molecule generation process. This approach aims to make the model use interpretable explanations from the XAI techniques to guide the optimization process, improve the quality of generated molecules, and enhance the exploration of the chemical space. The research will contribute to the development of a XAI enabled GAN architecture for molecule generation, which will be evaluated on various drug discovery tasks and compared to multiple state-of-the-art AI drug discovery models. Ultimately, this research seeks to address high failure rates of drug candidates, long timelines, and high costs in drug development by improving the reliability and effectiveness of AI-driven drug discovery.
Internal Reader: Dr. Ikjot Saini
External Reader: Dr. Andrew Swan
Advisor: Dr. Alioune Ngom
