Unveiling immune targets via TriStageHLA: high-throughput screening of peptide presentation, peptide-HLA binding, and immunogenicity

TriStageHLA predicts HLA class I epitope binding/presentation with AUROC 96.2% and AUPRC 95.2% on combined BA/EL evaluation, enabling efficient peptide prioritization.

• Why it matters: Proteome-scale epitope prioritization is hard because class I epitope discovery faces severe class imbalance, limiting reliable selection for neoantigen vaccines, T cell therapies, and immune monitoring.
• What they did: The three-stage framework uses PRE with a LoRA-adapted ESM-2 classifier, BIND with a dual-branch model integrating binding affinity and eluted ligand data, and IM fine-tuned on curated human T-cell assay data.
• The result: On held-out immunogenicity data, TriStageHLA-IM reaches AUROC 81.4% and AUPRC 73.8%, improving over ImmuneApp (70.7% / 63.6%) and BigMHC-IM (60.1% / 51.5%).

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