Machine Learning and Bioinformatics Models to Identify Gene Expression Patterns of Glioblastoma Associated with Disease Progression and Mortality

Glioblastoma (GB), the most frequent and aggressive form of malignant brain tumor, causes rapid death. The genetic mutation is the main reason for this fatal disease. Henceforth, it is essential to identify the causal genetic targets associated with GB survival. Plenty of publicly accessible gene expression and clinical data for GB patients from the Gene expression omnibus and the Broad Institute Cancer Genome Atlas (TCGA) dataset can allow us to study patient fatality prediction and thus to identify new GB biomarkers. In this study we applied bioinformatics and network-based approach to identify the altered genes associated with the GB comparing with normal mRNA expression data from the brain tissues. Total of 325 genes were found as differentially expressed in GB. Gene set enrichment analysis through protein-protein interaction (PPI), gene ontology (GO) and KEGG pathways also revealed their significance. We then applied machine-learning approach incorporating Cox Proportional Hazard models to the both clinical and RNA-Seq datasets to determine target biomarkers that affect the survival of the GB patients. Three genes (PHLDA1, IQGAP and, SPARC) were identified by using univariate approach that have a significant effect on the GB progression. Thus, our combined Machine Learning and Bioinformatics approach revealed the target signature genes for GB progression that could be useful to develop potential drug targets for the GB.