Integrated multi-omics analysis and machine learning refine molecular subtypes and clinical outcome for hepatocellular carcinoma.

The high morbidity and mortality of hepatocellular carcinoma (HCC) impose a substantial economic burden on patients' families and society, and the majority of HCC patients are detected at advanced stages and experience poor therapeutic outcomes, whereas early-stage patients exhibit the most favorable prognosis following radical treatment. In this study, we utilized a computational framework to integrate multi-omics data from HCC patients using the latest 10 different clustering algorithms, which were then employed a diverse set of 101 combinations derived from 10 different machine learning algorithms to develop a consensus machine learning-based signature (CMLBS). Using multi-omics consensus clustering, we distinguished two cancer subtypes (CSs) of HCC, and found that CS2 patients exhibited superior overall survival (OS) outcomes. In TCGA-LIHC, ICGC-LIRI, and multiple immunotherapy cohorts, low-CMLBS patients demonstrated favorable clinical outcomes and enhanced responsiveness to immunotherapy. Encouragingly, we observed that the high-CMLBS patients may exhibit increased sensitivity to Alpelisib, AZD7762, BMS-536,924, Carmustine, and GDC0810, whereas they may demonstrate reduced sensitivity to Axitinib, AZD6482, AZD8055, Entospletinib, GSK269962A, GSK1904529A, and GSK2606414, suggesting that CMLBS may contribute to the selection of chemotherapeutic agents for HCC patients. Therefore, in-depth examination of data from multi-omics data can provide valuable insights and contribute to the refinement of the molecular classification of HCC. In addition, the CMLBS model demonstrates potential as a screening tool for identifying HCC patients who may derive benefit from immunotherapy, and it possesses practical utility in the clinical management of HCC.