A Bioinformatics Analysis Based on Omics and Clinical Data for Graph-Based Patient Stratification in Hepatocellular Carcinoma

Hepatocellular carcinoma (HCC) is one of the most frequent and lethal malignancies worldwide, with substantial molecular and clinical heterogeneity that complicates prognostic assessment. We aimed to identify molecular subgroups with distinct prognostic profiles using a graph-based multi-omics approach. Similarity Network Fusion (SNF) was applied to integrate mRNA and miRNA sequencing data from the TCGA-LIHC cohort (366 patients), and spectral clustering on the fused similarity network was used to identify patient subgroups. Survival differences were assessed by log-rank test and Cox regression; subgroup characterisation combined direction-aware over-representation analysis and one-versus-all differential expression. Six clusters with distinct survival trajectories were identified in the discovery cohort (exploratory log-rank p-value = 3.6 × 10−4). C1 (24.3% of patients) showed the worst prognosis (median OS 22.6 months; 5-year OS 28.0%), while C4 reached the best outcomes (median OS not reached; HR 0.34, p-value = 0.001). Cluster membership retained independent prognostic value after adjustment for age, sex and AJCC stage. Each cluster was assigned a descriptive transcriptional label (proliferative–oncofoetal, xenobiotic-metabolising, imprinted/fetal-reactivated, hepatocyte-zonation-like, cholangio-mesenchymal, mature-hepatocyte). As independent evidence of prognostic relevance, the subtype structure was reproduced in a differently profiled external cohort (GSE14520; log-rank p-value = 0.011), with the poor-prognosis (C1) and favourable (C4) subtypes retaining their ordering. SNF-based integration provides a reproducible framework for prognostically informative patient stratification in HCC, with the molecular descriptors serving as starting points for downstream biological characterisation on independent cohorts.