Predicting clinical responses to checkpoint immunotherapy based on genetic and epigenetic alterations in cancer

Abstract

Cancer is one of the main causes of death in Korea and requires early detection and treatment. Accordingly, cancer immunotherapy using individual immune responses is emerging as a new treatment. In this study, we predicted clinical responses to immunotherapy largely based on genetic and epigenetic alterations. First, we constructed a convolutional neural network (CNN) based DeepNeo model to discover neoantigens that not only bind to major histocompatibility complexes (MHCs) but also react with T cells. As a result of predicting clinical responses to immunotherapy in multiple cancer types, it was confirmed that DeepNeo outperforms the previous prediction markers and consequently, we can expect a more accurate prediction with our model. Next, to predict responses based on methylation level, one of epigenetic changes, it was confirmed a high predictive power with selected 10 LINE-1 regions in tumor tissue samples. In addition, there was a substantial correlation between methylation levels of a cell free DNA (cfDNA) and of a tumor tissue. Finally, we suggested the probability that a cfDNA can be a non-invasive and simple prediction marker of cancer immunotherapy as an alternative for a tumor tissue.