EpiMII: Integrating Structure and Graph Neural Networks for MHC-II Epitope and Neoantigen Design

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• MHC-II neoantigens are crucial for immunotherapy, but accurate prediction remains challenging.
• Existing prediction tools for neoantigens have limited accuracy, resulting in low in vivo immunogenicity.
• Functional neoantigens derived from tumor DNA mutations are rare.

Purpose of the Study:

• To develop an accurate and efficient computational model for designing MHC-II epitopes.
• To improve the prediction of immunogenic neoantigens for cancer immunotherapy.
• To leverage structural features of epitopes for sequence prediction.

Main Methods:

• Development of EpiMII, a Graph Neural Network model for MHC-II epitope design.
• Construction of a large dataset of 142,934 MHC-II epitope structures for model training.
• Evaluation of EpiMII's performance against existing tools like ProteinMPNN.

Main Results:

• EpiMII achieved a 4.2x improvement over ProteinMPNN with a 78.0% sequence recovery rate for known MHC-II epitopes.
• Designed neoantigens from hepatocellular carcinoma activated CD4+ T cells in vitro, inducing IFN-γ and TNF-α secretion.
• One designed epitope significantly reduced tumor volume in mice, demonstrating in vivo efficacy.

Conclusions:

• EpiMII provides a novel and efficient method for identifying and designing MHC-II epitopes/neoantigens.
• The model's ability to predict functional epitopes holds significant potential for cancer vaccine development.
• This approach could overcome limitations of current neoantigen prediction tools, enhancing immunotherapy outcomes.