EpiMII: Structure-Aware Graph Neural Networks for MHC-II Epitope Generation

Major histocompatibility complex class II (MHC-II) neoantigens play a critical role in immunotherapy, either as direct effectors or through their influence on CD8⁺ T cells. However, only a small fraction of tumor DNA mutations qualify as functional neoantigens, and current prediction tools suffer from limited accuracy, making the in vivo design of highly immunogenic neoantigens challenging. Here, we present EpiMII, a structure-aware graph neural network model for de novo design of MHC-II epitopes. By integrating 3-dimensional structural information through an inverse folding strategy, EpiMII generates mimotopes that preserve T cell specificity while enhancing MHC-II binding affinity and immunogenicity. The model was trained on a first-of-its-kind curated dataset of 142,934 homology-modeled MHC-II epitope structures, overcoming the scarcity of experimentally resolved MHC-II-peptide complexes. Benchmarking analyses illustrate that EpiMII achieves a sequence recovery rate of 66.7% on a held-out test set and 79.0% on crystallized epitopes, markedly outperforming existing tools like ProteinMPNN. In a hepatocellular carcinoma case study, all 5 EpiMII-designed epitopes could markedly activate CD4⁺ T cells in vitro and induce the secretion of interferon-γ and tumor necrosis factor-α. Notably, treatment with one epitope (P4) substantially reduced tumor volume in mice in vivo, comparable to a known positive control. These findings establish EpiMII as a powerful tool for structure-guided neoantigen discovery, with broad implications for cancer vaccine development and personalized immunotherapy.