Abstract 1653: Modeling tumor immuno-dynamics to predict patient survival and immunotherapy efficacy

Abstract

Abstract State of the art models of cancer survival are increasingly utilizing molecular data that feature the number of non-synonymous mutations, or a network based stratification of mutations, as significant predictors. It is widely hypothesized that this association between mutations and survival is a result of neo-epitopes that induce a robust anti-tumor immune response. Naively this implies that patient neo-epitope burden should better predict survival than that of mutations alone, however this has generally been difficult to prove. Indeed, mutations and predicted neo-epitopes explain roughly the same survival variance, likely due to the intertwined difficulties of neo-epitope prioritization and the often immuno-suppressive tumor microenvironment. Reasoning that integrating leading order immune response with novel neo-epitope prioritization will lead to superior performance, we built a statistical model to quantify the influence of tumor immune-dynamics on patient survival. Neoepitopes (MHC-I/II) were predicted from mutations and filtered on a self-ligandome. The clonal structure of these mutations and the distribution of resulting epitopes were assessed as predictive features. Additionally, the existing response to epitopes was assayed by incorporating TCR and BCR sequence counts and entropy as a proxy for clonal tumor infiltrating lymphocyte expansion. Leveraging the cancer genome atlas (TCGA) for 9 major cancer types including bladder (BLCA), breast (BRCA), colorectal (COAD/READ), glioblastoma (GBM), liver (LIHC), lung (NSCLC), melanoma (SKCM), pancreatic (PAAD), and uterine (UCEC) cancer, we obtained omics data for a total of 2,866 patients. Random forest based recursive feature elimination was used to determine which features were most likely to be predictive while overfitting was controlled with k-fold cross validation. A cox proportional hazard model of survival as a function of mutation burden and functions of neoepitope and immune response predictors was constructed using the selected features. Our results confirm that neoepitope and immune response based predictors for survival often significantly outperform mutation burden alone and simultaneously suggest a quantitative classification of immunotherapy efficacy across cancer subtypes. Citation Format: Nicholas K. Akers, Eric E. Schadt, Bojan Losic. Modeling tumor immuno-dynamics to predict patient survival and immunotherapy efficacy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 1653. doi:10.1158/1538-7445.AM2017-1653