Single-cell and machine learning reveal ROS-associated heterogeneity in hepatocellular carcinoma for precision...

Reactive oxygen species (ROS) are critical regulators of tumor development and progression, yet their clinical significance and cellular mechanisms remain poorly defined in hepatocellular carcinoma (HCC).

A comprehensive computational approach was employed to investigate ROS-related heterogeneity in HCC and to develop a prognostic model. Transcriptomic analyses at the bulk and single-cell levels were integrated to assess cellular differentiation, metabolic activity, and intercellular signaling. Machine learning techniques were applied to construct a clinically interpretable ROS signature, which was validated through laboratory experiments.

Among the two identified molecular clusters, the high-ROS subtype was characterized by enhanced immunosuppressive cell infiltration, hyperactivation of oncogenic signaling pathways, and poorer clinical outcomes. scRNA-seq analyses revealed that malignant cells exhibit significantly elevated ROS levels. High ROS expression in tumor cells was associated with enhanced intercellular communication, metabolic reprogramming, and a progressive deterioration of malignant phenotypes. The constructed prediction signature could reliably forecast prognosis, outperforming 49 previously published models. Moreover, it accurately delineated genetic heterogeneity, anticipated responses to targeted medicines, and identified pertinent clinical conditions. Inhibition of PFKP markedly reduced the proliferation and migration of HCC cells while inducing apoptosis.

This study systematically delineates the molecular and cellular heterogeneity of ROS-associated subtypes in HCC, offering novel insights into precision medicine stratification and the development of targeted therapeutic strategies for ROS-driven malignancies.