Risk predictive model based on three immune-related gene pairs to assess prognosis and therapeutic sensitivity for hepatocellular carcinoma

Abstract

Background Hepatocellular carcinoma (HCC) as a common tumor has a poor prognosis. Recently, a combination of atezolizumab and bevacizumab has been recommended as the preferred regimen for advanced HCC. However, the overall response rate of this therapy is low. There is an urgent need to identify sensitive individuals for this precise therapy among HCC patients. Methods The Wilcox test was used to screen the differentially expressed immune-related genes by combining the TCGA cohort and the Immunology Database. Univariate and multivariate Cox regression analysis were used to screen the immune gene pairs concerning prognosis. A predictive model was constructed using LASSO Cox regression analysis, and correlation analysis was conducted between the signature and clinical characteristics. ICGC cohort and GSE14520 were applied for external validations of the predictive risk model. The relationship between immune cell infiltration, TMB, MSI, therapeutic sensitivity of immune checkpoint inhibitors, targeted drugs, and the risk model were assessed by bioinformatics analysis in HCC patients. Results A risk predictive model consisting of 3 immune-related gene pairs was constructed and the risk score was proved as an independent prognostic factor for HCC patients combining the TCGA cohort. This predictive model exhibited a positive correlation with tumor size (p < 0.01) and tumor stage (TNM) (p < 0.001) in the chi-square test. The predictive power was verified by external validations (ICGC and GSE14520). The risk score clearly correlated with immune cell infiltration, MSI, immune checkpoints, and markers of angiogenesis. Conclusions Our research established a risk predictive model based on 3 immune-related gene pairs and explored its relationship with immune characteristics, which might help to assess the prognosis and treatment sensitivity to immune and targeted therapy of HCC patients.

Machine learning-based investigation of regulated cell death for predicting prognosis and immunotherapy response in glioma patients

“…Previous predictive signatures for glioblastoma patient prognosis mainly relied on individual gene expression, disregarding the interactions or synergistic relationships between genes. In contrast, predictive models based on gene pairs offer several advantages including increased predictive power, enhanced interpretability and transferability to different datasets or experimental conditions 85 . Here, we comprehensively investigated the RCD landscape of glioblastoma from both bulk and single-cell perspectives.…”

Machine learning-based investigation of regulated cell death for predicting prognosis and immunotherapy response in glioma patients

“…Previous predictive signatures for glioblastoma patient prognosis mainly relied on individual gene expression, disregarding the interactions or synergistic relationships between genes. In contrast, predictive models based on gene pairs offer several advantages including increased predictive power, enhanced interpretability and transferability to different datasets or experimental conditions [42]. Here, we comprehensively investigated the RCD landscape of glioblastoma from both bulk and single-cell perspectives.…”