Deeplearning based MHC epitope prediction for cancer neoantigen discovery

Abstract

Neoantigens are important for cancer immunotherapies or cancer vaccine development, but identification of neoantigens is challenging. The high binding affinity between the mutated peptide and MHC (major histocompatibility complex) molecules of the patients is a necessary factor for a somatic mutation on the tumor genome to form a neoantigen. MHC epitope prediction tools can be used for the identification of neoantigens. This research investigates MHC epitope prediction by utilizing Tri-peptide similarity as features for the XGBoost classifier. This model was tested on experimentally validated cancer neoantigen peptides.