ECE PhD Prospectus Defense: Hao Yu

ECE PhD Prospectus Defense: Hao Yu

Title: Evaluating and Enhancing Docking- and Machine Learning–Based Methods for Structure-Based Epitope Mapping

Presenter: Hao Yu

Advisor: Professor Sandor Vajda

Chair: Professor Ioannis Paschalidis

Committee: Professor Sandor Vajda, Professor Ioannis Paschalidis, Professor Kayhan Batmanghelich, and Professor David Castañón

Google Scholar Link: https://scholar.google.com/citations?user=PCEWhbwAAAAJ&hl=en

Abstract: Epitopes are specific regions on antigens that are recognized and bound by antibodies. The process of predicting these sites, called epitope mapping or epitope prediction, is important for the development of antibody-based therapeutics. While experimental methods for epitope mapping exist, they often suffer from limitations in throughput and scalability. Computational epitope mapping offers a promising alternative for faster and more scalable discovery.

This study explores docking-based and machine learning-based approaches for epitope mapping. First, we demonstrate improvements in antibody-antigen docking by integrating the protein-protein docking algorithm ClusPro with AlphaFold2. This work lays the groundwork for a novel hybrid docking approach that combines physics-based modeling with machine learning. We propose to implement this novel approach to improve epitope mapping. Next, we evaluate existing docking-based and machine learning-based methods for epitope prediction. We found that machine learning methods using antigen latent representations from a protein language model as features showed strong performance in prediction accuracy and efficiency. To further improve the prediction specificity, we propose using a re-ranking strategy that incorporates antibody sequence information.

Overall, this prospectus presents an overview of current structure-based computational epitope mapping methods and proposes new strategies to enhance their accuracy and robustness.