基于DNA损伤修复基因构建AML预后预测模型

Construction of a prognostic prediction model for AML based on DNA damage repair genes

目的构建基于DNA损伤修复(DNA Damage Response,DDR)基因的急性髓系白血病(Acute Myeloid Leukemia,AML)预后模型,为探索AML基因稳定性和精准治疗提供理论依据。方法利用TCGA和GEO数据库分别下载训练集和验证集;利用MSigDB数据库的DDR通路基因,根据训练集中DDR基因表达水平与总生存(Overall survival,OS)相关系数,构建Lasso回归模型并得出特征基因;多因素COX回归得出每个基因的相关系数,并计算DDR评分(risk score);在训练集和验证集中分别根据DDR评分截断值,将2组病例分为高危组(high risk)和低危组(low risk),并对2组临床特征和预后进行比较;绘制受试者工作特征曲线(receiver operating characteristic curve,ROC)评价该DDR基因集的DDR评分预测OS的灵敏度和特异度。结果以TCGA(n=157)AML数据集作为训练集,建立Lasso回归模型后得到10个特征基因,将这10个特征基因合集命名为“DDR基因集”;高危组和低危组比较发现,高危组含ELN2017不良组(Adverse)病例比例更高(P<0.05);高危组OS较低危组明显缩短(P<0.05);年龄≥60岁(HR:2.853;95%CI:1.909~4.263),ELN2017危险度分层为不良组(HR:1.63;95%CI:1.059~2.511)以及高DDR评分(HR:3.137;95%CI:2.075~4.744)是影响OS的独立危险因素;验证集中发现高危组OS较低危组明显缩短(P<0.05);绘制TCGA数据集的ROC曲线显示:1年AUC=0.709,3年AUC=0.755和5年AUC=0.759。结论DDR基因集灵敏和稳健,DDR评分可用于临床预测AML的预后。

Objective To construct a prognostic model of Acute Myeloid Leukemia(AML)based on DNA damage response genes to provide a theoretical basis for exploring precision therapy in AML.Methods TCGA and GEO databases were used to download the training set and validation set,respectively.DDR pathway genes from the MSigDB database were downloaded,and the Lasso regression model was constructed based on the correlation coefficients between the expression level of DDR genes and OS in the training set,multifactorial COX regression was used to derive the correlation coefficients of each gene,and risk score was calculated.In the training set and validation set,the two groups of cases were divided into a high-risk group and a low-risk group according to the risk score cutoff value,and the clinical features and prognosis of the two groups were compared.ROC curves were plotted to evaluate the sensitivity and specificity of OS prediction by risk score of this DDR gene set.Results Using the TCGA(n=157)AML dataset as the training set,10 feature genes were obtained after Lasso regression modeling,and the 10 genes were named the“DDR gene set”.Comparison between the high-risk group and the low-risk group showed that the high-risk group contained a higher proportion of cases in the ELN2017 Adverse group(P<0.05).OS was significantly shorter in the high-risk group compared with the low-risk group(P<0.05).Age≥60 years(HR:2.853,95%CI:1.909~4.263)and ELN2017 Adverse subgroup(HR:1.63,95%CI:1.059~2.511),as well as high-risk score(HR:3.137,95%CI:2.075~4.744),were the independent risk factors affecting OS.The validation set found that OS in the high-risk group was significantly shorter than that in the low-risk group(P<0.05).ROC curves for the TCGA dataset showed that the one-year OS rate AUC=0.709,three-year OS rate AUC=0.755,and the 5-year OS rate AUC=0.759.Conclusion The DDR gene set is sensitive and robust and can be used for clinical prognosis of AML.