Integrative Machine-Learning Molecular Subtyping and Risk Scores of Circadian Rhythm-Related Prognostic Signatures in Colorectal Cancer.

PURPOSE: Colorectal Cancer (CRC) exhibits considerable heterogeneity. Circadian Rhythm (CR) disruption is increasingly implicated in tumorigenesis and cancer progression. METHODS: We identified Circadian Rhythm-Related Genes (CRRGs) significantly associated with prognosis through differential expression analysis and univariate Cox regression from 1,184 samples, and established the molecular subtypes based on unsupervised clustering. Employing a combination of ten machine learning algorithms to 101 model configurations, we developed and validated a high-predictive-performance risk-scoring model (CRRGscore). RESULTS: CRC patients were stratified into two distinct molecular subtypes (Cluster A vs. B). Cluster B had worse prognosis, and tumor microenvironment of Cluster B was characterized by enhanced immune suppression and stromal activation. The RSF model demonstrated the best performance (C-index = 0.707) and was used to build the CRRGscore. This risk model showed outstanding predictive ability in the TCGA training set, with 1-, 3-, 5-year AUCs of 0.982, 0.978, 0.991, respectively. Its robustness was maintained across three independent validation sets. Patients in the High-risk group had significantly poorer overall survival (P < 0.001) and the risk score was significantly correlated with adverse clinical characteristics, like advanced tumor stage and recurrence/metastasis (P < 0.01). Further analysis revealed an immune-excluded/immunosuppressive subtype phenotype in the High-risk group. Drug sensitivity analysis indicated that the High-risk group was less sensitive to conventional chemotherapeutics and targeted agents, and identifying several potential alternative drugs. CONCLUSION: This model demonstrated robust predictive capability for patient survival outcomes, tumor microenvironment status, and chemotherapeutic response in the integrated cohorts, providing a crucial clinical tool for prognostic assessment in CRC.