Construction and validation of a predictive model for the prognosis of hepatocellular carcinoma based on immunogenic cell death-related genes

Objective To construct a predictive model for the prognosis of patients with hepatocellular carcinoma (HCC) based on immunogenic cell death (ICD)-related genes, and to assess its clinical translational value. Methods A total of 365 HCC patients were selected from the Liver Hepatocellular Carcinoma (LIHC) cohort in The Cancer Genome Atlas database and were randomly divided into a training set and an internal validation set at a ratio of 7∶3, and a total of 108 HCC patients from the GSE76427 dataset in the Gene Expression Omnibus (GEO) database were selected as the external validation set. The bulk RNA-sequencing data of tumor tissue and the corresponding adjacent normal tissue were collected from the above patients, as well as related clinical data including sex, age, tumor grade, and overall survival. Single-cell RNA-sequencing data of 4 HCC patients were obtained from the GSE162616 dataset of the GEO database. Dimensionality reduction and clustering were performed for single cells from the GSE162616 dataset, and ICD activity scores were calcula-ted for each type of cells. Cells were divided into high ICD activity group and low ICD activity group according to ICD activity score, and differentially expressed genes (DEGs) between the two groups were identified. Weighted gene co-expression network analysis (WGCNA) was performed for the RNA-sequencing data of 365 HCC patients in the LIHC cohort to identify the gene set with the strongest association with ICD, and this gene set was intersected with the DEGs between HCC tumor tissue and the corresponding adjacent normal tissue of the patients in the LIHC cohort, finally obtaining the HCC-specific ICD-related genes (ICDRGs). The univariate Cox regression analysis was used to identify prognosis-related ICDRGs among the above genes; 10 machine learning algorithms were integrated to select the optimal genes for constructing an ICD-related HCC prognostic risk score model; the Spearman correlation analysis was used to investigate the correlation between the expression levels of these optimal genes and immune cell infiltration in HCC tissue. The model was used to calculate the prognostic risk score for each patient in the training set, the internal validation set, and the external validation set, and each set of patients were divided into high- and low-risk groups based on the median score. The Kaplan-Meier (K-M) survival analysis, the receiver operating characteristic (ROC) curve analysis, and the Cox regression analysis were used to evaluate the performance of the model in predicting prognosis. The univariate and multivariate Cox regression analyses were used to identify the influencing factors for the prognosis of HCC patients, which were integrated with the risk score to establish a nomogram model, and calibration curve and decision curve analysis (DCA) were used to assess the clinical translational value of the nomogram model. Results Five cell types were clustered in the GSE162616 dataset, among which immature B cells exhibited the highest ICD activity. A total of 920 ICD-related genes were identified at the single-cell level from the high and low ICD activity groups, and 82 HCC-specific ICDRGs were finally obtained through WGCNA. The univariate Cox regression analysis identified 26 prognosis-related ICDRGs from the above 82 genes, and after screening by machine learning algorithms, an ICD-related HCC prognostic risk score model was established based on 5 core genes, namely PIP4K2A, TAGLN2, RAC1, SLC16A3, and FYN. The Spearman correlation analysis showed that the expression levels of these 5 genes were significantly correlated with immune cell infiltration (P<0.001). The K-M survival analysis and the ROC curve analysis showed that this model had a good performance in predicting the prognosis of HCC patients, and the univariate and multivariate Cox regression analyses showed that the risk score of this model could be used as an independent predictive factor for the prognosis of HCC patients (HR=1.991,95%CI=1.368-2.898,P<0.001). The calibration curve showed that the nomogram model had a high accuracy in predicting the 1-, 3-, and 5-year survival rates of HCC patients, and DCA suggested that the nomogram model had the highest clinical benefit rate in terms of its performance in predicting 5-year survival rate. Conclusion The predictive model for the prognosis of HCC patients constructed in this study based on 5 HCC prognosis-related ICDRGs shows a good performance, and the nomogram model derived from the combination of this model and clinical indices also has a high clinical application value.