Multi-omics integration and machine learning define robust molecular subtypes and prognostic signatures in hepatocellular carcinoma.

backgroundHepatocellular carcinoma (HCC) exhibits high aggressiveness and substantial molecular heterogeneity, yet precise and individualized therapeutic strategies remain limited. Comprehensive multi-omics integration and machine learning-driven modeling hold promise for refining molecular subtyping and improving prognostic prediction in HCC.

methodsWe developed a computational framework integrating multi-omics datasets from HCC patients. Ten clustering algorithms were combined to perform integrative multi-omics clustering and identify molecular subtypes. Ten machine learning algorithms were subsequently applied to construct a consensus prognostic signature. The clinical relevance of subtypes and risk groups was assessed through survival analysis, immunotherapy response prediction, and tumor immune microenvironment profiling. Single-cell RNA sequencing and spatial transcriptomics data were incorporated to determine the cellular origins and spatial expression patterns of key genes.

resultsIntegrative multi-omics analysis identified four prognostically distinct cancer subtypes (CS1-CS4), with CS4 exhibiting the most favorable clinical outcomes. Five key genes were selected to build a robust prognostic model. Patients in the low-risk group demonstrated significantly better survival, an enhanced response to immunotherapy, and a higher probability of exhibiting a "hot tumor" phenotype. Conversely, the high-risk group showed poorer prognosis and reduced immunotherapy benefit. Single-cell and spatial transcriptomics analyses revealed that the key genes are predominantly enriched in malignant hepatocytes and display spatial patterns suggestive of tumor regional heterogeneity.

conclusionsThis study provides a refined molecular classification of HCC through integrative multi-omics analysis and establishes a machine learning-driven prognostic model with potential utility in early prognosis prediction and immunotherapy stratification. The dual-dimensional single-cell and spatial transcriptomic analyses further illuminate the cellular and spatial features associated with key prognostic genes. Prospective clinical validation is required to confirm the model's predictive performance.