FusionHLAII: A Hierarchical Model for Predicting HLA II–Peptide Interactions

In the field of immunotherapy for diseases, an accurate prediction of the binding of the HLA class II molecule to peptides is crucial for the development of effective vaccines and immunotherapies. We have developed a hierarchical deep learning model that innovatively combines amino acid physicochemical property encoding with advanced neural architectures, including the BiGRU and Transformer networks. The model captures temporal dependencies using BiGRU and leverages the multi-head self-attention mechanism in the Transformer to explore global interactions between HLA and peptides. Our model significantly improves the identification of key antigen-binding features, improving prediction accuracy and biological relevance. Experimental results demonstrate that, compared to existing methods, our approach shows significant advantages in both prediction performance and mechanistic understanding, offering more precise immune-binding predictions and providing a powerful computational tool and new perspectives for immunological research. Codes are available at https://github. com/LinsLab/FusionHLAII.