Mining key differentially expressed genes and construction of a prognostic risk score model for colorectal cancer...

Colorectal cancer (CRC) remains the third most common malignancy and the second leading cause of cancer-related death worldwide, with limited biomarkers available for accurate prognostic stratification. This study integrates multi-cohort transcriptomic data from four GEO datasets (GSE39582, GSE17536, GSE14333, GSE33113; n = 1123) and the TCGA-CRC cohort (n = 621) to identify key differentially expressed genes and construct a validated prognostic risk score model. Differential expression analysis using the limma framework identified 642 upregulated and 588 downregulated genes (|log2FC|>1, FDR<0.05). Weighted gene co-expression network analysis (WGCNA) extracted hub genes from modules significantly correlated with overall survival. LASSO Cox regression then reduced the hub gene set to an eight-gene prognostic signature (MMP1, AURKA, TOP2A, CDH1, COL1A1, CLDN1, ETV4, CDX2) that stratified patients into high- and low-risk groups with significantly different survival outcomes (p < 0.001, log-rank test). The model achieved AUCs of 0.824, 0.796, and 0.768 for 1-, 3-, and 5-year overall survival in the training cohort and was validated in three independent datasets. High-risk patients exhibited immunosuppressive microenvironments (elevated M2 macrophages, Tregs) and increased resistance to standard chemotherapy agents. Integration of the risk score with clinical variables improved prognostic accuracy (C-index 0.812). These findings offer a robust bioinformatics-derived prognostic tool with implications for immunotherapy patient selection and personalized treatment planning in CRC.