Development of molecular subtype specific prognostic marker signature in immune response associated colon cancer through fuzzy based transcriptomic approach

Abstract

Objective The molecular heterogeneity of colon cancer makes the prediction of disease prognosis challenging. In order to resolve this heterogeneity, molecular tumor subtyping present solutions. These approaches are expected to contribute to clinical decision-making. In this study, we aimed to identify Consensus Molecular Subtype (CMS) specific prognostic genes of colon cancer, focusing on anti-tumor immune- response associated CMS1, through a fuzzy-based machine learning approach. Materials and Methods We applied Fuzzy C-Means (FCM) clustering to stratify patients into two groups and identified genes that predict significant disease-specific survival difference between groups. We then performed Cox regression analyses to identify the most significant genes associated with disease-specific survival. A subtype-specific risk score and a final risk score formulae were constructed and used to calculate risk scores to stratify patients into low and high-risk groups within each CMS (1 to 4) or independent of CMS respectively. Results We identified CMS-specific genes and an overall 11-gene signature for prognostic risk prediction based on the disease-specific survival of colon cancer patients. The patients in both discovery and test cohorts were stratified into high and low-risk groups using subtype risk scores. The disease-specific survival of these risk groups within each CMS, except CMS3, was significantly different for both discovery and test cohorts. Discussion and Conclusions We have identified novel prognostic genes with potential immune regulatory roles within the immune-response associated CMS1. The low number of patients in the CMS3 cohort prevented subtype-specific prognostic gene validation. Tumor stage grouping of the validation cohort suggested the best prediction of prognosis in tumor stage III patients. In conclusion, newly identified eleven genes can efficiently predict the prognostic risk of colon cancer patients and classify patients into corresponding risk groups.