Rose, N. (BMEB) – MACHINE LEARNING MODELS FOR T CELL RECEPTOR TARGET DISCOVERY AND AFFINITY ENGINEERING

T cell receptors (TCRs) mediate antigen-specific immune responses through recognition of peptides presented by major histocompatibility complex (pMHC) molecules. Accurately predicting TCR–pMHC interactions remains a major barrier to TCR-based immunotherapy, due to limitations in current models that fail to generalize beyond common viral epitopes and well-characterized HLA alleles. In this PhD proposal, I outline a computational framework that integrates deep learning, structural modeling, and molecular simulation to improve TCR–pMHC prediction and discovery. I first introduce TRIOPS, a convolutional neural network trained on harmonized, experimentally validated data to predict MHC restriction from TCR sequence alone. TRIOPS outperforms state-of-the-art models in both held-out evaluation and independent patient datasets, demonstrating improved accuracy in assigning TCRs to their correct restricting MHC alleles. I then propose TRILOBITE, a two-part structure-based model combining atomic-resolution graph representations and dynamics-derived biophysical features to classify TCR–pMHC binding and estimate affinity. Finally, I propose an end-to-end pipeline to identify tumor-reactive TCRs from patient-derived sequencing data by integrating HLA typing, antigen prediction, structure generation, and binding assessment. Applied to a pan-cancer atlas of over 1.2 million T cells, this framework will enable high-throughput, structure-informed TCR discovery across diverse HLA backgrounds. Together, these aims address a critical need for scalable, mechanistically grounded methods for mapping T cell specificity to accelerate cancer immunotherapy discovery.

Event Host: Nicholas Rose, PhD student, Biomolecular Engineering & Bioinformatics

Advisor: Vanessa Jonsson