HyperPMT: MHC-Peptide-TCR Binding Prediction via UniGAT-Based Hypergraph Neural Network

Accurately predicting the binding of major histocompatibility complex (MHC)-peptide-T cell receptor (TCR) ternary complexes is essential for advancing tumor immunotherapy and T-cell therapy development. Current approaches predominantly focus on binary interactions or rely on two-stage prediction strategies that fragment the holistic biological process into independent steps. Consequently, these methods struggle to capture the high-order interplay among the three components within a unified framework. Addressing this constraint, we introduce HyperPMT, an innovative approach based on the UniGAT hypergraph neural network for predicting MHC-peptide-TCR ternary complex binding. HyperPMT models CDR3β sequences, antigenic peptides, and HLA pseudo-sequences as three distinct node types within a hypergraph, where each complete ternary complex is represented as a hyperedge connecting three nodes. Through a two-stage message passing mechanism, HyperPMT learns high-order node representations, which enables direct, end-to-end inference of binding outcomes. Comparative evaluations reveal that HyperPMT achieves markedly superior performance over current methods on key evaluation metrics, including ROC-AUC and PR-AUC. HyperPMT introduces a new technical pathway for predicting MHC-peptide-TCR ternary interactions and holds potential to provide computational support for tumor immunotherapy and personalized vaccine design.