Deep mapping of structural perturbations to energetics enables precise TCR design.

Abstract

Affinity optimization and cross-reactivity profiling are pivotal for T cell receptor (TCR) engineering but remain hindered by repertoire diversity and limited structural insights. We present mpTCRai, a deep learning framework that predicts residue-level structural interactions by simulating the sequential mechanics of antigen presentation and T cell recognition. Through a hotspot-based scoring mechanism, mpTCRai explicitly maps structural perturbations to energetic changes, revealing that these perturbations strongly correlate with experimental affinities (r = -0.88). The model captures the structural dependencies of key molecular switches and computationally screens against cross-reactive mutations like Y5W to evaluate structural off-target risks. Guided by these insights, we computationally prioritized four A6-TCR variants for adult T cell leukemia. Overall, this work establishes a unified computational platform integrating structural and energetic constraints to accelerate the downstream experimental validation and rational design of engineered TCRs.