Integrative Multi-omics and Supervised Learning Identifies an Epithelial Signature for Radiotherapy Response in Colorectal Cancer

Colorectal cancer (CRC) is the third most diagnosed cancer globally, accounting for 9.6% of all cancer cases and the second leading cause of cancer deaths. One of the common treatments for the disease is radiotherapy, which can come with dangerous side effects and varying outcomes in patients. Such factors highlight the need for personalized treatments, that depend on identifying biomarkers capable of effectively predicting patient responses to radiotherapy. Here, we integrated four data types including RNA, copy number alteration (CNA), mutational profiles and methylation profiles using a multi-omics factor analysis (MOFA), towards identifying predictive markers associated with therapeutic response. We identified a panel of 101 biomarkers associated with patients responses to radiotherapy that were further validated using a random forest classifier on an independent testing cohort. Our panel demonstrated effectiveness in predicting treatment outcomes, with an 89% accuracy to differentiate patients with complete response to radiotherapy compared to non-complete responders. We have demonstrated that this panel is predominantly expressed in CRC epithelial cells. This study presents a combined unsupervised and supervised approach to underscore multiple patterns of variation associated with treatment outcomes, and presents a predictive panel of marker genes associated with radiotherapy treatment response in CRC.