HEPATOCELLULAR carcinoma recurrence was predicted with high accuracy using a transcriptome-based machine learning model, while genetic analyses also identified COL2A1 as a potential protective factor associated with disease risk and recurrence related biological processes.
Postoperative recurrence remains a major challenge in hepatocellular carcinoma, prompting researchers to investigate molecular markers that could improve risk prediction. Transcriptomic data were integrated from three independent datasets containing 756 samples to identify genes associated with recurrence. This analysis identified 141 differentially expressed genes linked to postoperative recurrence.
Researchers evaluated 113 machine learning strategies, including standalone models and combinations of feature selection methods with classifiers. A random forest (RF) model demonstrated the strongest overall performance and identified 24 core genes that formed the basis of the recurrence prediction model. Bootstrap optimism correction supported the stability of the model, while exploratory external validation in an independent dataset showed continued predictive performance.
The RF model achieved an average area under the curve of 0.930 and maintained favourable performance during external validation with an area under the curve of 0.889. Mendelian randomisation analysis suggested that increased expression of COL2A1 was associated with a lower risk of hepatocellular carcinoma (odds ratio: 0.811; 95% CI: 0.669–0.984; p=0.034).
Single cell RNA sequencing demonstrated higher COL2A1 expression in hepatocytes, macrophages, endothelial cells, and tissue stem cells from samples without microvascular invasion compared with samples showing microvascular invasion. These findings suggested that COL2A1 may be associated with biological features linked to reduced tumour aggressiveness.
Functional enrichment and immune infiltration analyses indicated that COL2A1 may influence extracellular matrix remodelling, epithelial mesenchymal transition, and regulation of the tumour immune microenvironment. These pathways have been associated with recurrence and tumour invasion, providing further biological context for the observed genetic findings.
The investigators concluded that the 24 gene RF model demonstrated promising performance for classifying postoperative hepatocellular carcinoma recurrence. They also suggested that COL2A1 may contribute to recurrence associated microenvironmental states through immune regulation and extracellular matrix remodelling. However, they emphasised that further validation is required before the prediction model or the potential use of COL2A1 as a biomarker can be considered for clinical application.
Reference
Kong Y et al. Identification of therapeutic targets and postoperative recurrence prediction model construction for hepatocellular carcinoma based on systematic mendelian randomization. Arab J Gastroenterol. 2026;S1687-1979(26)00068-7.
Featured Image: karrastock on Adobe Stock
Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.
