A Singapore team of clinician-scientists and researchers has developed a machine-learning tool to predict which liver cancer patients are likely to experience recurrence after surgery.
The tool combines genomic and clinical information and was found to outperform the commonly used tumour, node and metastasis staging system, which assesses disease stage primarily based on tumour burden.
Researchers from the National Cancer Centre Singapore, Duke-NUS Medical School and the A*STAR Genome Institute of Singapore also identified two biologically distinct ways in which hepatocellular carcinoma, the most common form of primary liver cancer, can recur.
The findings could help clinicians tailor follow-up care, identify patients who may benefit from additional treatment after surgery and support the design of more targeted clinical trials.
The study was conducted through the National Medical Research Council-funded PLANet research programme, or Precision Medicine in Liver Cancer across an Asia-Pacific Network, and was published in Gut in July 2026.
Hepatocellular carcinoma frequently returns after surgical resection, but clinicians currently lack reliable biomarkers to determine which patients face the greatest risk. Intrahepatic recurrence, where the cancer returns within the liver, accounts for 70% to 80% of recurrences, while the remaining cases involve distant metastasis.
The research team analysed clinical data and conducted matched genetic analyses of tumour samples from patients in the PLANet cohort, including recurrent tumour samples from a subset of patients.
Among the 106 patients studied, 68, or 64.2%, experienced recurrence. This included 48 patients whose cancer returned within the liver, 11 whose cancer recurred elsewhere in the body and nine who experienced recurrence in both locations.
The researchers identified two main recurrence patterns. Polyclonal seeding occurs when multiple groups of cancer cells spread simultaneously from the original tumour and was more commonly associated with recurrence within the liver. Tumours in this group displayed characteristics suggesting that they may respond better to certain immunotherapies.
Monoclonal seeding occurs when recurrence develops from a single clone originating from the primary tumour. This pattern was more commonly associated with later recurrence and cancer spreading beyond the liver.
“By uncovering two distinct mechanisms of liver cancer recurrence, we have gained new insights into the biological processes that drive disease progression after surgery,” said Dr Zhang Ying, co-first author of the study and Senior Scientist at A*STAR GIS.
“Understanding these differences is an important step towards more personalised approaches to risk prediction, and the development of more precise biomarkers and targeted therapies for HCC patients.”
Building on the findings, the team developed a machine-learning-based multi-omics tool that predicts a patient’s risk of liver cancer recurrence after surgery.
The tool combines tumour size and genetics, blood-marker levels, cancer stage and a group of 15 genes strongly associated with recurrence.
It was validated across three independent cohorts, including The Cancer Genome Atlas Liver Hepatocellular Carcinoma dataset. The dataset comprises a largely Western population and complements the Asia-Pacific population represented in the PLANet programme.
The tool achieved a predictive performance score of 86%, compared with 56% to 68% for TNM staging.
“The tool we developed is robust and comprehensively evaluated,” said Karthik Sekar, co-first author of the study and Senior Bioinformatics Specialist at NCCS, who led its development.
“With it, we can now determine which patients are most likely to experience recurrence and, by extension, most likely to benefit from specific systemic therapies, ensuring that they are neither overtreated nor undertreated.”
Professor Pierce Chow, co-senior author of the study and Principal Investigator of the PLANet programme, said the tool could support the selection of patients for studies involving adjuvant therapies intended to prevent recurrence.
The research team is now using spatial sequencing technologies to identify tumour-microenvironment biomarkers that could serve as drug targets for hepatocellular carcinoma. It is also working to improve the multi-omics tool by incorporating additional data sources, including CT imaging.
