MHC-I epitope prediction via repertoire-wide TCR-peptide contact profiles

Improved prediction of immunogenicity of major histocompatibility complex class I (MHC-I) ligands, also known as epitopes, could aid the development of effective vaccine and tailored immunotherapy in infection and cancer. Several attempts to discern epitopes from non-immunogenic MHC ligands from peptide-MHC interaction profiles have so far yielded limited success. Here, we demonstrated that epitopes can be accurately identified from repertoire-wide TCR-peptide contact profiles parametrized by only one contact potential scale. The optimized immunogenicity prediction model demonstrated overall accuracy and area under the curve (AUC) of 0.75 and 0.82, respectively. Predictive performance was not biased for epitopes loaded on 5 major HLA supertypes. Position-specific effects of TCR-peptide interactions on immunogenicity were revealed. Our analysis demonstrated that TCR-peptide interaction is a major determinant of immunogenicity. Prospective validation in vaccination and cancer immunotherapy cohorts is warranted.