Prioritizing candidate peptides for cancer vaccines by PEPPRMINT: a statistical model to predict peptide presentation by HLA-I proteins

Abstract

Recent development of cancer immunotherapy has opened unprecedented avenues to eliminate tumor cells using the human immune system. Cancer vaccines composed of neoantigens, or peptides unique to tumor cells due to somatic mutations, have emerged as a promising approach to activate or strengthen the immune response against cancer. A key step to identifying neoantigens is computationally predicting which somatically mutated peptides are presented on the cell surface by a human leukocyte antigen (HLA). Computational prediction relies on large amounts of high-quality training data, such as mass spectrometry data of peptides presented by one of several HLAs in living cells. We developed a complete pipeline to prioritize neoantigens for cancer vaccines. A key step of our pipeline is PEPPRMINT (PEPtide PResentation using a MIxture model and Neural neTwork), a model designed to exploit mass spectrometry data to predict peptide presentation by HLAs. We applied our pipeline to DNA sequencing data of 60 melanoma patients and identified a group of neoantigens that were more immunogenic in tumor cells than in normal cells. Additionally, the neoantigen burden estimated by PEPPRMINT was significantly associated with activity of the immune system, suggesting these neoantigens could induce an immune response.