低级别胶质瘤自噬相关基因预后分类器的构建及验证

目的本研究旨在构建自噬相关基因(ARGs)的风险分类器，从而预测低级别胶质瘤（LGG) 患者的生存率。方法从UCSC Xena, CGGA, GlioVis和GTEx公共数据库中获取LGG患者和正常脑组织数据，结合人类自噬数据库筛选出232个ARGs。通过差异分析得到差异ARGs。在训练集中，利用单因素Cox回归分析和LASSO回归分析，构建ARGs的预后分类器。通过绘制受试者工作特征（ROC）曲线确定最佳Cut-off值。Kaplan-Meier 生存曲线和ROC曲线下面积（AUC）用于评估分类器性能，并在内部数据集和外部数据集中验证。Cox回归分析用于评估分类器的独立预后价值。最后，结合常见临床参数和风险分类用于构建列线图模型，并利用ROC曲线，一致性指数和校准曲线评估该模型的预测能力。结果本研究构建了由6个ARGs(BAG1、 PTK6、 EEF2、 PEA15、 ITGA6和 MAP1LC3C)的预后风险分类器，其可将LGG患者分为具有明显生存差异的高、低风险组在多个数据集(均P＜0.05 )。5年AUC值显示该分类器在训练集，内部验证集和TCGA总集中分别为0.837, 0.755, 和0.803。同时，在外部验证集中，1年，2年和3年ROC曲线依然提示该分类器具有较好的预测准确性。Cox回归分析显示该预后分类器在来自TCGA的多个数据集中都具有独立预后价值(HR>1,P<0.05)。之后构建了包含多个常见临床参数和预后风险分类的列线图模型，时间依赖性ROC，一致性指数(C-index)和校准曲线分析同时证实了该模型的预后价值。结论本研究建立了一个由6个ARGs的具有高预后价值的风险分类器，并结合临床病理特征和风险分类构建了用于临床决策的列线图，能更好地帮助临床医生判断LGG患者的预后和进行个体化治疗的临床决策。

Objective To construct a risk classifier for autophagy related genes(ARGs) to predict the survival rate of low-grade glioma(LGG) patients. Methods LGG patient and normal brain tissue data were obtained from the public databases of UCSC Xena, CGGA, GlioVis, and GTEx, and screen 232 ARGs using a human autophagy database. Differential ARGs were obtained through differential analysis. In the training set, a prognostic classifier for ARGs was constructed using univariate Cox regression analysis and LASSO regression analysis. The optimal Cut-off value was determined by drawing receiver operating characteristic(ROC) curves. The Kaplan Meier survival curve and area under the ROC curve(AUC) were used to evaluate classifier performance and validated on both internal and external datasets. Cox regression analysis was used to evaluate the independent prognostic value of classifiers. Finally, a column chart model was constructed using common clinical parameters and risk classification, and the predictive ability of the model was evaluated using ROC curves, consistency indices, and calibration curves. Results This study constructed a prognostic risk classifier consisting of six ARGs(BAG1, PTK6, EEF2, PEA15, ITGA6, and MAP1LC3C), which can classify LGG patients into high-risk and low-risk groups with significant survival differences across multiple datasets(all P<0.05). The 5-year AUC value shows that the classifier has values of 0.837, 0.755, and 0.803 in the training set, internal validation set, and TCGA total set, respectively. Meanwhile, in the external validation set, the ROC curves for 1 year, 2 years, and 3 years still indicated that the classifier has good predictive accuracy. Cox regression analysis showed that the prognostic classifier had independent prognostic value in multiple datasets from TCGA(HR>1,P<0.05). Afterwards, a column chart model containing multiple common clinical parameters and prognostic risk classification was constructed, and the prognostic value of the model was confirmed through time dependent ROC(t-ROC), consistency index(C-index), and calibration curve analysis. Conclusion A risk classifier with high prognostic value is established consisting of 6 ARGs, and combined clinical pathological features and risk classification to construct a column chart for clinical decision-making, which can better assist clinical doctors in judging the prognosis of LGG patients and making personalized treatment clinical decisions.