PredIG: an interpretable predictor of T-cell epitope immunogenicity.

Area of Science:

• Immunology and Bioinformatics
• Computational Biology and Drug Discovery

Background:

• Cytotoxic T cells are crucial for fighting pathogens and tumors.
• Identifying immunogenic T-cell epitopes is vital for antigen-based immunotherapies.
• Current methods for epitope discovery face challenges in throughput and accuracy.

Purpose of the Study:

• To develop an interpretable predictor for T-cell epitope immunogenicity.
• To improve the success rate of immunogenicity screening for T-cell epitopes.
• To identify key drivers of T-cell epitope immunogenicity beyond HLA-I binding affinity.

Main Methods:

• Trained PredIG on 17,448 peptide-HLA-I allele pairs with reported immunogenicity.
• Integrated in silico features including antigen processing, HLA-I binding, and physicochemical properties.
• Utilized XGBoost models and SHAP for prediction and interpretability analysis.

Main Results:

• PredIG achieved cutting-edge immunogenicity screening success rates (ISSR) for pathogen and non-canonical cancer antigens.
• Analysis revealed balanced feature importance between antigenic and physicochemical properties.
• Prioritized additional immunogenic neoantigens beyond top-binding candidates.

Conclusions:

• PredIG scores refine and expand the prioritization of T-cell epitopes for infection and cancer immunotherapies.
• PredIG offers unprecedented immunological interpretability, highlighting drivers beyond HLA-I binding.
• PredIG facilitates high-throughput antigen discovery via open-source tools and a webserver.