General prediction of T cell receptor antigen specificity from sequence using AlphaFold 3.

AlphaFold 3 predicts MHC:peptide:TCR triad structures with unprecedented accuracy, achieving median AUCs of 0.81–0.92 on validation triads that were unseen during training.

• Why it matters: Decoding TCR antigen specificity from sequence is a bottleneck, and existing in silico methods struggle to predict binding to unseen epitopes.
• What they did: The study applied AlphaFold 3 to >9000 TCRs mapped to >1000 epitopes across >70 MHC class I/II alleles, then learned sequence-linked features separating cognate triads from controls.
• The result: The model’s performance on validation triads unseen by AlphaFold 3 or during feature selection suggests generalized TCR specificity prediction from sequence is feasible.

AI summary of the abstract · may contain errors · not medical advice · read the paper.