Molecular Dynamics and Computational Predictions of T-cell Activation

Predicting T-cell receptor (TCR) specificity is one of the hardest challenges in immunology owing to difficulties in perturbing this system experimentally and computationally. Successful prediction of TCR specificity can offer a powerful to tool to study T-cell tolerance, antigen recognition, and autoimmunity as well as open doors towards de-novo TCR design and TCR-based Chimeric Antigen Receptor (CAR) T therapies. In this work, we developed a computational model based on experimental insights, molecular dynamics simulations (MD), and machine learning (ML) that can reliably predict TCR specificity towards peptide-class II major histocompatibility (pMHCII) complexes. Our model leverages biophysical information from MD simulations of AlphaFold3 generated TCR-pMHCII complexes to reliably predict T-cell activation outcomes. Inclusion of a novel biophysical parameter termed nonconvex interactions combined with interaction energy calculations, allowed us to improve T-cell activation predictions over an ML model only trained on interaction energies and also offer a compelling theory for TCR specificity encoding via kinetic entrapment.