Currently, cancer stem cells (CSCs) are regarded as the most promising target for cancer therapy due to their close association with tumor resistance, invasion, and recurrence. Thus, identifying CSCs-related genes and constructing a prognostic risk model associated with CSCs may be crucial for hepatocellular carcinoma (HCC) therapy. Xena Browser was used to download gene expression profiles and clinical data, while MSigDB was used to obtain genes associated with CSCs. Firstly, the non-negative matrix factorization (NMF) algorithm was used to cluster the HCC samples based on CSCs-related genes. To evaluate the predictive performance of the risk model, the receiver operating characteristic curves (ROC) and Kaplan–Meier analysis were used. The R package “rms” was used to construct the final nomogram based on risk scores and clinical characteristics. Based on 449 CSCs-related genes, a total of 588 HCC samples from TCGA-LIHC and ICGC-LIRI_JP were classified into four molecular subtypes with marked differences in survival and mRNA stemness index (mRNAsi) between subtypes. Univariate Cox regression, multivariate Cox regression, and LASSO regression analyses were performed on a total of 1417 differentially expressed genes (DEGs) between subtypes, and a nine-gene prognostic model was constructed with TTK, ST6GALNAC4, SPP1, SGCB, MEP1A, HTRA1, CD79A, C6, and ATP2A3. In both the training and testing sets and the external validation cohort, the risk model performed well in predicting HCC patients’ survival. A nomogram was constructed and had high predictive efficacy in short-term survival. In comparison with the other two prognostic models, our nine-gene signature model performed best. We constructed a nine-gene signature model to predict the survival of HCC patients, which has good predictive efficacy and stability. The model may contribute to guiding the prognostic assessment of HCC patients in clinical practice.

目前，癌症干细胞（CSC）由于与肿瘤抵抗、侵袭和复发密切相关，被认为是最有希望的癌症治疗靶点。因此，识别 CSC 相关基因并构建与 CSC 相关的预后风险模型可能对于肝细胞癌 (HCC) 治疗至关重要。Xena Browser用于下载基因表达谱和临床数据，而MSigDB用于获取与CSC相关的基因。首先，使用非负矩阵分解（NMF）算法根据CSC相关基因对HCC样本进行聚类。为了评估风险模型的预测性能，使用了受试者工作特征曲线（ROC）和卡普兰-迈耶分析。R 包“rms”用于根据风险评分和临床特征构建最终列线图。基于449个CSC相关基因，TCGA-LIHC和ICGC-LIRI_JP总共588个HCC样本被分为四种分子亚型，亚型之间的生存率和mRNA干性指数（mRNAsi）存在显着差异。对总共1417个亚型间差异表达基因（DEG）进行单变量Cox回归、多变量Cox回归和LASSO回归分析，并利用TTK、ST6GALNAC4、SPP1、SGCB、MEP1A、HTRA1、 CD79A、C6 和 ATP2A3。在训练和测试集以及外部验证队列中，风险模型在预测 HCC 患者的生存方面表现良好。构建了列线图，并且对短期生存具有很高的预测功效。与其他两个预后模型相比，我们的九基因特征模型表现最好。我们构建了九基因特征模型来预测HCC患者的生存，具有良好的预测效果和稳定性。该模型可能有助于指导临床实践中HCC患者的预后评估。