Integrative predictive modelling and multi-omics validation reveals potential indicators of clinical heterogeneity in HCC.

Background

The escalating incidence and mortality rates of hepatocellular carcinoma (HCC), coupled with delayed diagnosis, resistance to existing therapies, and profound molecular heterogeneity, underscore the urgent need for improved, clinically actionable biomarkers.

Methods

Large-scale transcriptomic datasets were analysed to identify differentially expressed genes in HCC, followed by pathway enrichment and protein-protein interaction network analyses to derive robust hub genes. An integrative predictive modelling and explainable artificial intelligence (AI) approach was used to classify HCC tumours and prioritise candidate biomarkers. Clinical significance was evaluated using survival modelling, while their multi-omics validation incorporated hepatocyte cell line models, immunohistochemistry profiles, single-cell expression landscapes, immune infiltration patterns, and genomic alteration frequencies. Their expressions were further mapped to different HCC aetiologies, and drug-gene interaction networks were constructed.

Results

The best-performing XGBoost model exhibited strong discriminative performance on external tumour tissue datasets, while reliably predicting the tumour cases in serum-derived exosomal samples within a limited cohort. Five biomarker candidates were prioritised based on predictive importance and multi-omics relevance. CDK1, CDC20, and E2F1 were consistently overexpressed in malignant hepatocytes with protein-level and clinical support, confirming their pre-existing roles as oncogenic mitotic regulators. ADRA1D showed stromal-endothelial enrichment, indicating a previously unrecognised microenvironment-linked role, while GLP2R exhibited tumour-suppressive properties with consistent downregulation, representing newer potential context-specific roles in HCC. These genes showed variable expression patterns in cross-cohort analyses across different aetiologies and also depicted association with multiple drugs.

Conclusions

This integrative framework identified and mechanistically characterised five potential HCC biomarkers with utility in precision oncology, requiring real-world benchmarking.