NanoCMSer: a consensus molecular subtype stratification tool for fresh‐frozen and paraffin‐embedded colorectal cancer samples

Abstract

Colorectal cancer (CRC) is a significant contributor to cancer‐related mortality, emphasizing the need for advanced biomarkers to guide treatment. As part of an international consortium, we previously categorized CRCs into four consensus molecular subtypes (CMS1‐CMS4), showing promise for outcome prediction. To facilitate clinical integration of CMS classification in settings where formalin‐fixed paraffin‐embedded (FFPE) samples are routinely used, we developed NanoCMSer, a NanoString‐based CMS classifier using 55 genes. NanoCMSer achieved high accuracy rates, with 95% for fresh‐frozen samples from the MATCH cohort and 92% for FFPE samples from the CODE cohort, marking the highest reported accuracy for FFPE tissues to date. Additionally, it demonstrated 96% accuracy across a comprehensive collection of 23 RNAseq‐based datasets, compiled in this study, surpassing the performance of existing models. Classifying with only 55 genes, the CMS predictions were still biologically relevant, recognizing CMS‐specific biology upon enrichment analysis. Additionally, we observed substantial differences in recurrence‐free survival curves when comparing CMS2/3 patients in stage III versus II. Probability of recurrence after 5 years increased by 21% in CMS2 and 31% in CMS3 for patients in stage III, whereas this difference was less pronounced for CMS1 and CMS4, with 11% and 10%, respectively. We posit NanoCMSer as a robust tool for subtyping CRCs for both tumor biology and clinical practice, accessible via nanocmser r package (https://github.com/LEXORlab/NanoCMSer) and Shinyapp (https://atorang.shinyapps.io/NanoCMSer).

Keywords

• colorectal cancer
• consensus molecular subtypes
• machine learning
• NanoString
• prognosis biomarker