Decoding Glioblastoma Treatment Responses: A Combined Bioinformatics and Machine Learning Study

Glioblastoma (GBM), or grade IV astrocytoma, is the most intense and prevalent type of gliomas, which also causes over ۶۰% of brain tumors. Even with the major improvements in the cancer treatments, GBM patients still have a dismal prognosis and a very poor survival rate. In this study, using a combination of bioinformatics and machine learning methods, our purpose is to examine the differentially expressed genes and identify important molecular pathways and biomarkers incorporated in the response to various GBM treatments. Materials and methods: In this study, we utilized the R TCGAbiolinks package to download the TCGA-GBM dataset classified into five groups: normal samples (n = ۵), patients who received pharmaceutical therapy such as chemotherapy (group ۱, n = ۵), patients who received radiotherapy (group ۲, n = ۶), patients who received both treatments (group ۳, n = ۱۳۳), and patients who received no treatment (group ۴, n = ۱۶). To find differentially expressed genes (DEGs), we used the R DESeq۲ package with an adjusted p value of less than ۰.۰۵. Following that, we carried out Gene Set Enrichment Analysis (GSEA) to identify enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and Gene Ontology (GO) biological processes. To distinguish between patients who received both treatments and those who received neither, we applied feature selection and classification methods, including Random Forest (RF) and Extreme Gradient Boosting (XGBoost) with the help of python libraries. Results: RPLP۰P۶ was the most differentially expressed gene among all the groups; DLG۲, RPSAP۵۸, RRM۲, and H۳F۳AP۴ were the next differentially expressed genes for groups ۱, ۲, ۳, and ۴, respectively. In terms of KEGG pathways, the glutamatergic synapse pathway was significantly enriched in both the patients who received pharmaceutical therapy, and radiotherapy, as well as those who did not receive pharmaceutical or radiation therapy. Additionally, GO pathways such as synaptic membrane, postsynaptic membrane, and neuron to neuron synapse were significantly enriched in group ۳. Among classification methods, RF outperformed XGBoost, with an area under the curve (AUC) of ۰.۸۱, using OSMR, FAM۲۲۸A, TSSC۲, PIEZO۱P۲, PDE۶C, LINC۰۱۶۴۵, FOXE۱, MIR۳۱۴۲HG, SSTR۵-AS۱, and RAB۳۸ as the top ۱۰ important features. Conclusion: Using DESeq۲, GSEA, and machine learning techniques, we found molecular signatures and potential biomarkers relevant to treatment responses in GBM patients. The Random Forest model demonstrated superior predictive performance, highlighting potential biomarkers for further investigation in therapeutic strategies.

کلیدواژه ها:

Glioblastoma (GBM) ، Bioinformatics ، Machine Learning ، Random Forest (RF) ، Biomarkers