STEP: Prediction of TCRbeta-epitope binding specificity via structure-informed physicochemical interaction maps - Data and Model

Data and trained model supporting the manuscript STEP: Prediction of TCRβ–epitope binding specificity via structure-informed physicochemical interaction maps, submitted to Protein Science. Contents: epitope-disjoint train/test splits (600 training epitopes; 581 unseen test epitopes, n = 3,314 balanced), cancer-specific evaluation sets, precomputed five-channel physicochemical interaction maps, consensus contact-probability priors derived from 198 experimentally determined TCR–pMHC complexes (1.40–3.00 Å), and the trained model checkpoint (best_model.pt). Code: github.com/Imrans-AI/STEP