Differentiation of glioblastoma multiforme solid and recurrent tumors using gene expression profile and machine learning algorithms

The most prevalent type of brain tumor is called grade IV glioma, or glioblastoma multiforme(GBM). GBM is among the deadliest tumors, with an extremely poor survival rate. GBM patients havea ۵-year survival rate of only ۵% [۱]. In the present research, we used bioinformatics methods todiscover a group of genes that demonstrate notable differential expression between GBM patients andnormal samples. Also, we applied machine learning algorithms to find potential biomarkers for GBMpatients. The R language program "TCGAbiolinks" was used to download the TCGA-GBM transcriptsper million (TPM) data comprising ۱۷۵ samples: ۱۳ recurrent glioblastomas, ۱۵۷ primary tumors, and۵ solid tissue normal samples. Using an adjusted p-value of less than ۰.۰۵, the top ۱۰ differentiallyexpressed genes between GBM and normal samples in the TCGA-GBM cohort were selected with theR programming package "DESeq۲." The LASSO algorithm, also known as the Least AbsoluteShrinkage and Selection Operator, was utilized for feature selection, along with other methods such asrandom forest classifier (RFC), support vector classifier (SVC), and artificial neural networks (ANNs)to distinguish between the datasets. The differentially expressed top ۱۰ genes (MAPK۹, MIGA۱,WDR۷, CDKL۵, C۲CD۲L, AAK۱, AKAP۵, UBR۳, CPEB۳, and SYNJ۱) were screened by DESeq۲analysis. LASSO analysis led to the identification of ۱۰ genes (ANK۳, GCNT۴, CD۱۷۷P۱, LINC۰۲۹۷۴,CDKL۵, GAS۷, ADRB۳, KLHL۲, LINC۰۲۵۰۰, and LDHAL۶EP) as the feature selections linked tothe prognosis of glioblastoma. The area under the curve (AUC) of ANNs, RFC, and SVC using rawdata was ۰.۴۹, ۰.۹۴, and ۰.۹۱, as well as ۰.۶۹, ۰.۹۴, and ۰.۰۸ using selected features, respectively.

کلیدواژه ها:

Glioblastoma Multiforme ، gene expression ، machine learning ، DESeq۲ ، LASSO