构建基于20基因的急性髓系白血病预后生存模型

Construct a prognostic survival model of acute myeloid leukemia based on 20 genes

目的利用Firehose数据库的数据,分析与急性髓系白血病(AML)发生、发展及预后相关的分子标志物,构建基于机器学习算法的AML 1年预后模型。方法从GDC(Genomic Data Commons)的外部链接Broad Firehose数据库中下载关于AML患者的临床及转录组数据,筛选出符合要求的生存期及mRNA测序数据的病历共163例。运用R语言DESeq程序包进行差异表达基因的筛选,并应用R语言的Rattle程序包构建基于20基因的AML 1年预后生存模型。结果 EBF4、MTUS2、NT5E、AEF2、IGDCC4基因表达水平上调,与预后良好有关,ADAMTS2、TRPM4、PACSIN1、CACNG4、SPON1、CCDC3、C10orf72、MAOA、ESPN、C1QA、LILRA4、UBXN10、LIF、WDR86、PEG10基因表达水平上调,与不良预后相关,可以作为AML发生、发展的相关生物标志物。与决策树(Desicion Tree)、随机森林(RF)、支持向量机(SVM)、线性回归(Linear Regression)、人工神经网络(ANN)AML预后模型相比,Boost模型曲线下面积(AUC)值最高,为0.75。结论基于机器学习算法构建的模型能较准确地预测AML的预后,Boost预后模型判断AML患者1年预后的预测效果更佳。

Objective To analyze the molecular markers related to the occurrence,development and prognosis of acute myeloid leukemia(AML)and to construct a 1-year prognostic model of AML based on machine learning algorithm using the data of Firehose database.Methods Clinical and transcriptome Data of AML patients were downloaded from the Broad Firehose database of GDC(Genomic Data Commons).A total of 163 patients with the survival time and mRNA sequencing Data meeting the requirements were screened out.The Deseq program of R language was used to screen the differentially expressed genes,and the Rattle program of R language was used to construct a 1-year prognostic survival model of AML based on 20 genes.Results The up-regulating expression levels of EBF4,MTUS2,NT5 E,AEF2,IGDCC4 genes were associated with good prognosis,while the up-regulating expression levels of ADAMTS2,TRPM4,PACSin1,CACNG4,Spon1,CCDC3,C10 ORF72,MAOA,ESPN,C1 qa,LILRA4,UBXN10,LIF,WDR86 and PEG10 genes were associated with poor prognosis,and could be used as biomarkers for the occurrence and development of AML.Compared with Desicion Tree,Random Forest(RF),Support Vector Machine(SVM),Linear Regression(Linear Regression)and Artificial Neural Network(ANN)AML prognostic models,the Area Under Curve(AUC)of the Boost model was the highest(0.75).Conclusion The model based on machine learning algorithm can predict the prognosis of AML patients more accurately,and the Boost prognostic model has a better prediction effect in predicting the 1-year prognosis of AML patients.