Machine Learning-based Gene Biomarker Identification for Improving Prognosis and Therapy in Hepatocellular Carcinoma

Purpose: Traditional clinical evaluations based on pathological classification have shown limited effectiveness in predicting prognosis and guiding treatment for patients with hepatocellular carcinoma (HCC). This study aims to identify a robust molecular biomarker for improving prognosis and therapy in HCC. Methods: The International Cancer Genome Consortium (ICGC), Gene Expression Omnibus (GEO), and The Cancer Genome Atlas (TCGA) provided expression data and clinicopathological information for several cohorts. . First, Cox regression analysis and differentially expressed analysis were performed to identify robust prognostic genes. Next, machine learning algorithms, including 101 statistical models, were employed to pinpoint key genes in HCC. Single-cell sequencing analysis was conducted to explore the potential subcellular functions of each key gene. Based on these findings, an HCC Prognosis- Related Index (HPRI) was developed from the identified key genes, and HPRIbased nomogram models were validated across multiple cohorts. Additionally, tumor microenvironment analysis and drug sensitivity analysis were performed further to assess the clinical significance of the HPRI in HCC. Results: A total of 36 robust prognostic genes with differential expression in HCC were identified, from which seven key genes-DCAF13, EEF1E1, GMPS, OLA1, PLOD2, PABPC1, and PPARGC1A-were filtered using machine learning algorithms. Except for PPARGC1A, all these genes were highly expressed in malignant cells, followed by fibroblasts. Notably, we developed the HPRI based on the key genes and validated its clinical relevance. Results demonstrated that the HPRI and HPRI-derived nomogram models had good predictive performance across multiple cohorts. Following tumor microenvironment analysis revealed that a high HPRI was linked to a higher likelihood of immune evasion. Drug sensitivity analysis suggested that patients with a high HPRI might benefit from chemotherapeutic agents like sorafenib, as well as novel compounds such as ML323 and MK-1775. Conclusion: Our study demonstrates a well-rounded approach by integrating gene expression, machine learning, tumor microenvironment analysis, and drug sensitivity profiling. HPRI may serve as a promising predictor for guiding prognosis and personalized treatment in HCC. result: A total of 36 robust prognostic genes with differential expression in HCC were identified, from which seven key genes-DCAF13, EEF1E1, GMPS, OLA1, PLOD2, PABPC1, and PPARGC1A-were filtered using machine learning algorithms. Except for PPARGC1A, all these genes were highly expressed in malignant cells, followed by fibroblasts. Notably, we developed the HPRI based on the key genes and validated its clinical relevance. Results demonstrated that the HPRI and HPRI-derived nomogram models had good predictive performance across multiple cohorts. Following tumor microenvironment analysis revealed that a high HPRI was linked to a higher likelihood of immune evasion. Drug sensitivity analysis suggested that patients with a high HPRI might benefit from chemotherapeutic agents like sorafenib, as well as novel compounds such as ML323 and MK-1775.