Written by Lauren Robertson, Science Writer
As the third leading cause of cancer-related death worldwide, liver cancer represents a significant challenge to both researchers and healthcare providers. Hepatocellular carcinoma (HCC) accounts for 90% of all cases and exhibits a high mortality and morbidity rate. Patients at the same stage of disease can present with different molecular subtypes, and it is vital to correctly identify these subtypes to improve prognosis and ensure effective treatment.
In a study published in Frontiers in Genetics, researchers used a machine learning approach based on multi-kernel learning to integrate key data types across the omics disciplines and help elucidate three HCC molecular subtypes.
Managing multi-omics data
In the context of cancer and other diseases, data from different omics technologies can give an insight into how a tumour forms and how it is progressing. But individual omics only reveal part of the picture. To get a more complete view of cancer, researchers are increasingly turning to multi-omics data integration.
Multiple-kernel learning is one example of a machine learning approach that can be used to integrate such diverse datasets. In the current study, researchers turned to a specific approach known as rMKL-LPP (regularised multiple kernel learning with locality preserving projections) that performs dimensionality reduction and data integration simultaneously.
rMKL-LPP was used to integrate mRNA, miRNA and DNA methylation data of 287 HCC patients. In total, 2 subtypes were identified that showed significantly different 3-year mortality rates – the high-risk group with a 51% mortality rate and the low-risk group with 23.5%. Overall, the high-risk patients were 3.37 times more likely to die compared to the low-risk group.

Elucidating the biological processes of HCC
After homing in on the distinct subtypes, the authors wanted to get a better understanding of the biological processes underlying this disease. Further analysis highlighted 6 pathways that were significantly different between the two group. Specifically, the activity of the Hypoxia, MAPK, EGFR, NF-kβ, and TNFα pathways were found to be higher in the high-risk group, while the low-risk group exhibited higher VEGF pathway activity. As the authors state: “This suggests that pathway-blocking therapy can provide new opportunities for precise treatment of HCC.”
Immune cell infiltration analysis also revealed that 9 immune cells seemed to be present at differing concentrations across the two subtypes. In the high-risk group, monocytic lineage, CD8+T cell, T cell, myeloid dendritic cell, and cytotoxicity score was found to be significantly higher than in the low-risk group. These tumour-infiltrating immune cells could potentially prove to be potent targets for drug delivery systems and other therapeutic strategies in HCC.

Lastly, weighted gene co-expression network analysis identified various gene modules that may impact prognosis. Importantly, these genes were found to be involved in key biological processes that may aid the development of liver cancer, helping to elucidate the mechanisms underlying HCC progression. Hub gene analysis showed that high expression levels of CDK1, CDCA8, TACC3, and NCAPG were significantly associated with poor HCC prognosis, and could be used as biomarkers in the future.
Though further validation of these results is needed, “the selected potential pathogenic genes, pathways and tumour-infiltrating immune cells can be used as references to control related gene expression or interfere with their target signal transduction pathways to provide potential opportunities for the treatment of HCC,” say the study authors. “Our findings may bring novel insights into the subtypes of HCC and promote the realization of precision medicine.”