Increasing evidence has highlighted the biological significance of 5-methylcytosine (5mC) DNA modification in regulating tumorigenesis and cancer progression. However, the potential roles of the 5mC modification in HCC are still unknown. A consensus clustering algorithm was performed to determine 5mC modification patterns and establish a 5mC-related gene signature in HCC. Weighted gene co‑expression network analysis (WGCNA) integrated with machine learning approaches (LASSO, RF, and SVM‑RFE) was applied to identify candidate biomarkers, and a 5mC risk score model was subsequently constructed. The concordance index (C‑index), decision curve analysis (DCA), and ROC curve were employed to evaluate the performance of the nomogram. Two distinct 5mC modification patterns were identified, with cluster B exhibiting worse prognosis and enrichment of cell cycle and metabolic pathways. Using integrated WGCNA and machine learning, we constructed a three-gene (DNMT1, SPATS2, MCM6) risk score model that effectively stratified patients into high- and low-risk groups with significantly different overall survival. The risk score demonstrated robust prognostic performance across independent cohorts and was an independent prognostic factor. High risk scores correlated with advanced stage, higher tumor mutational burden, distinct immune microenvironment features, and increased sensitivity to multiple chemotherapeutic and targeted agents. A nomogram incorporating the risk score and clinicopathological features showed acceptable predictive accuracy and clinical net benefit. Experimental validation confirmed upregulation of DNMT1 and SPATS2 in HCC tissues, and their knockdown suppressed HCC cell proliferation, migration, and invasion. The 5mC risk model demonstrates clinical applicability and net benefit in terms of prognostic stratification, immunophenotype characterization, and prediction of drug therapy response in HCC patients, providing a useful tool for precision oncology.
Hepatocellular carcinoma (HCC) has a dismal prognosis and profound metabolic reprogramming, yet the role of glycolysis-associated genes (GRGs) remains incompletely defined. This study aimed to develop a glycolysis‑associated prognostic model and explore its association with the tumor immune microenvironment (TME). We curated 760 GRGs from MSigDB and performed consensus clustering to identify molecular subtypes. Using WGCNA integrated with three machine learning algorithms (LASSO, Random Forest, and SVM-RFE), we constructed a GRGs risk score model and validated it across multiple independent cohorts. Its correlation with prognosis, genomic alterations, immune infiltration, and in silico drug sensitivity was comprehensively evaluated. Experimental validation was conducted in clinical specimens and HCC cell lines. Two distinct GRGs-based subtypes were identified, with cluster B exhibiting worse prognosis and significant enrichment of cell cycle, glycolysis, EMT, and PI3K‑AKT‑mTOR pathways. A four-gene risk model (CBX1, DNMT1, GAS2L3, KPNA2) was constructed. The risk score demonstrated moderate prognostic performance as an independent prognostic factor and correlated with advanced AJCC stage, higher tumor mutational burden, and an immunosuppressive TME characterized by increased Tregs and M0 macrophages. A nomogram incorporating the risk score showed moderate predictive accuracy and clinical net benefit. Experimental validation supported upregulation of all four genes in HCC tissues, and knockdown of CBX1 or GAS2L3 markedly suppressed cell proliferation, migration, and invasion in vitro. The GRGs risk model demonstrates clinical applicability and net benefit in terms of prognostic stratification and immunophenotype characterization in HCC patients, offering a potentially useful tool for risk assessment and personalized management.