Identification of potential diagnostic hubgenes for endometriosis using meta-analysis and machinelearning approaches

Background: Endometriosis is a benign gynecological disorderwhich is identified as precursor lesion for several malignancies.This study can help to understand the molecular mechanismsof endometriosis.Materials and Methods: Four gene expression datasets(GSE۷۳۰۵, GSE۷۳۰۷, GSE۲۵۶۲۸, GSE۱۱۶۹۱) were obtainedfrom GEO. After preprocessing steps with normalized unscaledstandard error (NUSE) and relative log expression (RLE) methods,the remaining samples were normalized using GCRMA.Three datasets (GSE۷۳۰۵, GSE۷۳۰۷, GSE۲۵۶۲۸) were mergedinto one and batch effects were adjusted using ComBat function.Those genes with |log۲ fold-change (FC)|>۳ and adjustedp-value<۰.۰۱ were considered to be differentially-expressed(DEGs) using limma package. MiRNAs and transcription factors(TFs) which regulate the DEGs were obtained from miRWalkand TRRUST database and a TF-mRNA-miRNA networkwas built using Cytoscape. Hub genes were identified based onfour algorithms including degree, betweenness, closeness andMCC and were used to build a diagnostic logistic regressionmodel for endometriosis.Results: After quality assessment using RLE and NUSE, threeand four samples were unqualified in GSE۷۳۰۵ and GSE۷۳۰۷,respectively so they were excluded from the analysis. Aftercombining three datasets and batch effect removal, ۱۵۳ geneswere considered to be DEGs. Next, total of ۳۲۴ miRNA–mRNA and ۱۰۹ TF-mRNA pairs were combined to form a TF–miRNA–mRNA network. Five hub genes (GATA۶, HMOX۱,HS۳ST۱, NFASC, PTGIS) were detected using the intersectionbetween top ten mRNAs based on four different algorithms. Formodel construction, a logistic regression model was establishedusing the merged dataset as training set. The AUC of the modelwas ۰.۹۸ and ۰.۸۵ on the training and validation (GSE۳۰۶۰۱)sets respectively, using ۵-fold cross-validation.Conclusion: In the current study, utilizing machine learning approacheswe identified a set of five genes, which can discriminatebetween endometriosis and non-endometriosis samples

کلیدواژه ها:

Endometriosis ، meta-analysis ، machine learning ، diagnostic