Molecular classification and construction of prognostic risk model via machine learning based on metabolic-related
Hongkang Wang1, Yonghao Li2, Huayu Zhang3
1Faculty of Science and Engineering, The University of Manchester, Manchester, UK.
Background:
Despite recent advances in therapeutic strategies for hepatocellular carcinoma (HCC), patient prognosis remains unsatisfactory. Accumulating evidence indicated that dysregulation of metabolism-related pathways and genes plays a pivotal role in HCC progression. Accordingly, metabolic-associated genes hold promise as prognostic biomarkers and potential predictors of therapeutic response. This study aimed to identify distinct metabolic subtypes of HCC and compare their clinicopathological and genomic features. Prognosis-associated metabolic genes were screened using an integrated machine learning framework, from which a risk-scoring model was constructed to simultaneously predict HCC prognosis and immunotherapy response.
Methods:
A systematic evaluation of metabolic patterns was conducted to elucidate the association between metabolism and HCC. To establish a prognostic and therapeutic prediction model, we developed an integrative framework based on machine learning algorithms by using comprehensive profiling of RNA sequencing data. Subsequently, nine metabolic-related genes were identified, and their biological functions were validated with clinical characteristics, immune cell infiltration, and complicated cellular signaling pathways.
Results:
Two molecular clusters with distinct clinical and biological characteristics were identified in HCC. Utilizing the computational framework, a metabolic-based prognostic model was constructed, which exhibited superior prognostic accuracy and outperformed previously reported models. An efficient clinical nomogram integrating the risk score with clinicopathological variables was subsequently established. Metabolic status was found to be closely associated with immunological features, and the proposed algorithms effectively predicted immunotherapy responsiveness. Furthermore, the risk score demonstrated predictive power for drug sensitivity in HCC patients. A multilevel evaluation of prognosis-related metabolic genes confirmed the stability and robustness of the model.
Conclusions:
This study systematically demonstrated the relationship between metabolic alterations and HCC. We established a robust prognostic model which was capable of accurately predicting patient survival, prognosis, and therapeutic responses. This model held promise for improving clinical decision-making and advancing personalized treatment strategies in HCC.
