Machine learning–driven approaches for predicting T-cell–mediated immunity and beyond.

Abstract

2588 Background: Recognition of peptides presented by the major histocompatibility complex (MHC) through the T cell receptor (TCR-pMHC) is crucial for T cell function, influencing disease conditions such as cancer, infections, and autoimmune disorders. Despite previous attempts, predictive models of TCR-pMHC specificity remain challenging. Methods: Inspired by recent breakthroughs in protein structure prediction achieved by deep neural networks, we explored structural modeling using AlphaFold 3 (AF3)-based AI-enabled computation as a potential avenue for predicting TCR epitope specificity. Results: We show that a specialized version of the neural network predictor AlphaFold can generate models of TCR-pMHC interactions, effectively distinguishing valid peptide epitopes from invalid ones with increasing accuracy. Strongly immunogenic epitopes could be identified and selected for vaccine development through in-silico high-throughput processes. Higher-affinity and specificity T cells could also be computationally designed to achieve improved efficacy and safety profiles for T cell therapy. An accurate TCR-pMHC prediction model is expected to significantly benefit T-cell-mediated immunotherapy and facilitate advanced drug design. Conclusions: Overall, precise prediction of T-cell immunogenicity holds substantial therapeutic potential, enabling the identification of peptide epitopes associated with tumors, infectious agents, and autoimmune diseases. Although much work remains before these predictions could achieve widespread practical utility, deep learning-based structural modeling represents a promising path toward the generalizable predictions of TCR-pMHC interactions and beyond.