The researchers introduced ImmunoStruct, a novel deep learning model designed to accurately predict the immunogenicity of peptide–MHC complexes for vaccine development. Unlike traditional tools that rely solely on amino acid sequences, this architecture integrates multimodal data, including 3D structural insights from AlphaFold2 and specific biochemical properties. To address the complexities of cancer neoantigens, the model employs a contrastive learning strategy that differentiates between mutant and wild-type peptides. Results demonstrate that ImmunoStruct outperforms existing methods in identifying effective epitopes across both infectious diseases and various cancers. Furthermore, the model’s predictions correlate strongly with patient survival outcomes and have been validated through in vitro assays using SARS-CoV-2 samples. This framework provides an interpretable approach to immunotherapy design by highlighting the specific molecular regions most critical for T cell activation.

References:

• Givechian K B, Rocha J F, Liu C, et al. ImmunoStruct enables multimodal deep learning for immunogenicity prediction[J]. Nature Machine Intelligence, 2025: 1-14.