Exercise-related immune gene signature for hepatocellular carcinoma: machine learning and multi-omics analysis.

Background: Exercise is known to regulate the immune system. However, its prognostic value in hepatocellular carcinoma (HCC) remains largely unknown.

Objective: This study aims to construct a machine learning-based prognostic signature using exercise-related immune genes (EIGs) to predict prognosis in HCC.

Methods: We obtained mRNA-seq and scRNA of HCC from GeneCards, GEO, TCGA and ICGC. EIG were obtained using WGCNA, differential gene expression analysis and CIBERSORT. Univariate COX analysis and 101 combinations of 10 machine learning algorithms were used to construct EIG prognostic signature (EIGPS), and survival analyses were performed. Furthermore, we conducted molecular subtyping, qRT-PCR, biological functions, immune infiltration, drug sensitivity, and single cell analyses on EIGPS.

Results: Using WGCNA, differential gene expression analysis, and CIBERSORT, 59 EIGs were identified, of which 54 were associated with prognosis. EIGPS constructed by 7 EIGs (UPF3B, G6PD, ENO1, FARSB, CYP2C9, DLGAP5, SLC2A1) had the highest average C-index value (0.742), showing good predictive performance independent of clinical features. qRT-PCR results showed that CYP2C9 was lowly expressed in HCC cells, while all other genes were highly expressed. 7 EIGs were divided into two subtypes, with C2 exhibiting better anti-tumor immunity. Immunological biological differences between high- and low-risk groups based on EIGPS involved immune responses. EIGPS was mainly expressed in macrophages. The high-risk group had higher macrophage abundance and immune escape ability, as well as greater sensitivity to Afatinib and Alpelisib.

Conclusions: We identified key EIGs and constructed an EIGPS that can effectively predict the prognosis of HCC, which offers avenues for better personalized treatments.