A Machine-Learning Prognostic Model for Colorectal Cancer Using a Complement-Related Risk Signature.

OBJECTIVES: Colorectal cancer (CRC) remains a major contributor to global cancer mortality, ranking second worldwide for cancer-related deaths in 2022, and is characterized by marked heterogeneity in prognosis and therapeutic response. We sought to construct a machine-learning prognostic model based on a complement-related risk signature (CRRS) and to situate this signature within the CRC immune microenvironment. METHODS: Transcriptomic profiles with matched clinical annotations from TCGA and GEO CRC cohorts were analyzed. Prognostic CRRS genes were screened using Cox proportional hazards modeling alongside machine-learning procedures. A random survival forest (RSF) predictor was trained and externally validated. Comparisons of immune infiltration, mutational burden, pathway enrichment, and drug sensitivity were made between risk groups. The function of FAM84A, a key model gene, was examined in CRC cell lines. RESULTS: The six-gene CRRS model accurately stratified patients by survival outcomes. Low-risk patients exhibited greater immune cell infiltration and higher predicted response to immunotherapy and chemotherapy, while high-risk patients showed enrichment of complement activation and matrix remodeling pathways. FAM84A was shown to promote CRC cell proliferation, migration, and epithelial-mesenchymal transition. CONCLUSION: CRRS is a critical modulator of the CRC immune microenvironment. The developed model enables precise risk prediction and supports individualized therapeutic decisions in CRC.