LAUSR

632Background: Colon cancer is highly heterogeneous in prognosis and response to treatment. The consensus molecular subtypes (CMS1-4, and mixed) partition colon cancers into distinct groups. CMS4 tumors, a mesenchymal subtype, have the worst prognosis and poor response to standard chemotherapies. There is a critical need for accurate molecular subtyping, and subtype-specific management. Methods: Affymetrix microarrays colon cancer datasets (N = 813; GSE39582, GSE14333) were partitioned into training (AT; N = 370) and validation sets (AV; N = 443) balanced for clinical traits. A novel multistate gene methodology was used to predict CMS, and prognosticate subtype-specific relapse-free survival (RFS) in the training set. Accuracy of CMS prediction and prognostic significance was validated in the AV and TCGA colon cancer (COAD; N = 458) sets. Results: In the training set, a 20-gene panel (ColotypeR-CMS) predicts CMS subtype. Mean accuracy for CMS1-4 prediction was 0.87 in AV and 0.81 in COAD. In AV, 5-year RF...