Machine Learning Diagnostic Model for Hepatocellular Carcinoma Based on Liquid-Liquid Phase Separation and Ferroptosis-Related Genes.

Background/aims: Hepatocellular carcinoma (HCC) represents a primary liver malignancy with a multifaceted molecular landscape. The interplay between liquid-liquid phase separation (LLPS) and ferroptosis-a regulated form of cell death-has garnered interest in tumorigenesis. However, the precise role of LLPS and ferroptosis-related genes in HCC progression and prognosis remains obscure. Unraveling this connection could pave the way for innovative diagnosis and therapeutic strategies.

Materials and methods: The differentially expressed genes (DEGs) were identified based on 3 GEO datasets, followed by overlapping with LLPS-related and ferroptosis-related genes. Based on central hub genes, a diagnostic model was developed through LASSO regression and validated using KM survival analysis and real-time quantitative polymerase chain reaction (RT-qPCR). Then the effects of NRAS on the development of HCC and ferroptosis were also detected.

Results: We identified 24 DEGs overlapping among HCC-specific, LLPS, and ferroptosis-related genes. A diagnostic model, centered on 5 hub genes, was developed and validated. Lower expression of these genes corresponded with enhanced patient survival rates, and they were distinctly overexpressed in HCC cells. NRAS downregulation significantly inhibited HepG2 cell proliferation and migration (P < .01). Fe2+ content and ROS levels were both significantly increased in the si-NRAS group when compared to those in the si-NC group (P < .01), while opposite results were observed for the protein level of GPX4 and GSH content.

Conclusion: The diagnostic model with 5 hub genes (EZH2, HSPB1, NRAS, RPL8, and SUV39H1) emerges as a potential innovative tool for the diagnosis of HCC. NRAS promotes the carcinogenesis of HCC cells and inhibits ferroptosis.