Kevin Ma
Research Mentor: Yongqun He
Mentor Department: Not Available, Medicine
Author(s): Kevin Ma, Yichao Chen, Yongqun He
Session: Session 4 (1:00 PM – 1:50 PM)
Presentation Type: Poster 52
Abstract
Advances in machine learning have significantly strengthened reverse vaccinology–based vaccine discovery. In this work, we introduce protein language model–derived representations into vaccine antigen prediction, marking the first integration of protein folding–based features in this context. Using Evolutionary Scale Modeling (ESM), we extracted 2,560 numerical features from protein sequences and assessed their effectiveness across multiple learning setups. We find that ESM features alone provide predictive performance comparable to manually engineered features, and their combination leads to further, though modest, improvements. Evaluation under a leave-one-pathogen-out validation scheme demonstrates enhanced antigen prediction for 10 bacterial pathogens. This metho is also being used to predict vaccine antigens from Mycobacterium tuberculosis, the bacterium that causes tuberculosis.
