Integrative analysis of DNA methylation and gene expression to identify gastric cancerdiagnostic biomarkers via machine learning approache

Gastric cancer is the second cause of cancer-related deaths. Most patients are diagnosed in a late stage becausethere are not proper methods for early cancer detection. Therefore, there is an urgent need for finding suchbiomarkers for this cancer. In the current study, our purpose is to integrate methylation and gene expressiondata from The Cancer Genome Atlas (TCGA) to find potential diagnostic biomarkers for gastric cancer viabioinformatics and machine learning approaches. DNA Methylation and gene expression data for gastriccancer were downloaded from TCGA using TCGABiolinks R-package. At first, probes locating at sexchromosomes, containing single nucleotide polymorphisms (SNPs) or missing values were removed. Then,we used ChAMP R-package for finding differentially methylated CpGs (DMCs). CpGs with adjusted pvalues<۰.۰۵ and |delta β| > ۰.۲۵ were considered DMCs. Gene expression dataset was normalized viaDESeq۲ R-package. Genes were considered differentially expressed (DEGs) if they satisfied the thresholdof |log۲ fold change| > ۱ and adjusted p-values<۰.۰۵. Since promoter hyper-methylation of tumor suppressorgenes is one of the most important observations in cancer, we only continued with hyper-methylated CpGs(۳۸ probes) located in the promoter of downregulated genes. Recursive feature elimination with crossvalidation(RFECV) method was used to find features with highest discriminative power between tumoraland normal samples resulting in ۴ final probes including cg۱۰۶۰۴۶۴۶, cg۲۲۰۸۳۰۴۷, cg۰۷۷۳۰۳۲۹ andcg۱۲۷۴۱۴۲۰. These features where then used for constructing a logistic regression model. We validated thesemarkers in an independent set from GEO database (GSE۳۰۶۰۱). The area under the curve (AUC) of modelwas ۰.۹۰۴ indicating that the four markers could achieve excellent performance in distinguishing tumoraland normal gastric samples. Overall, the four high-performance diagnostic signatures built through machinelearning approaches can improve gastric cancer precision management upon prospective clinical validation.

کلیدواژه ها:

Machine learning ، DNA methylation ، Gene expression ، Diagnosis ، Gastric cance