Abstract

Effective cancer immunotherapy hinges on the precise recognition of neoantigens, presented as binary complexes with major histocompatibility complex (MHC) molecules, by T cell receptors (TCR). The development of immunogenic peptide predictors and generators plays a central role in personalizing immunotherapies while reducing experimental costs. However, the current methods often fall short in leveraging structural data efficiently and providing comprehensive guidance for neoantigen selection. To address these limitations, we introduce NEOM, a novel neoantigen maturation framework encompassing five distinct modules: "policy", "structure", "evaluation", "selection" and "filter". This framework is designed to enhance precision, interpretability, customizability and cost-effectiveness in neoantigen screening. We evaluated NEOM using a set of random synthetic peptides, followed by available clinically-derived peptides. NEOM achieved higher performance on generated peptide quality compared to other baseline models. Using established predictors for filtering revealed a substantial number of peptides with immunogenic potential. Subsequently, a more rigorous binding affinity evaluation using free energy perturbation methods identified 6 out of 38 candidates showing superior binding characteristics. MHC tetramer peptide exchange assays and flow cytometry experiments further validate five of them. These results demonstrate that NEOM not only excels in identifying diverse peptides with enhanced binding stability and affinity for MHC molecules but also augments their immunogenic potential, showcasing its utility in advancing personalized immunotherapies.