Machine learning dataset and benchmark for accurate T cell receptor-pHLA binding prediction.

Hi-TPH-PLMs fine-tuned on Hi-TPH achieve a 19.2% performance gain over state-of-the-art tools on an external wet-lab test set for TCR–pHLA binding prediction.

• Why it matters: Accurate T cell receptor–peptide-human leukocyte antigen interaction prediction is hindered by incomplete sequence data and biased nonbinding examples, limiting immunology and therapeutic design.
• What they did: The study introduces Hi-TPH, a hierarchical dataset with on-the-fly selection for generating nonbinding data, and fine-tunes multiple protein language models (PLMs) on it.
• The result: Hierarchical analyses show component contributions and synergy, including that HLA prediction contribution relies on full T cell receptor chains, and the released datasets enable more reliable immunoinformatics and personalized immunotherapy research.