A 63 signature genes prediction system is effective for glioblastoma prognosis

The present study aimed to explore possible prognostic marker genes in glioblastoma (GBM). Differentially expressed genes (DEGs) were screened by comparing microarray data of tumor and normal tissue samples from The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) dataset GSE22866. Subsequently, the prognosis-associated DEGs were screened via Cox regression analysis, followed by construction of gene/protein/pathway interaction networks of these DEGs by calculating the correlation coefficient between the DEGs. Next, a prognostic prediction system was constructed using Bayes discriminant analysis, which was validated by the microarray data of samples from patients with good and bad prognosis from the TCGA and Chinese Glioma Genome Atlas (CGGA), as well as the GEO dataset. Finally, a co-expression network of the signature genes in the prediction system was constructed in combination with the significant pathways. A total of 288 overlapping DEGs (false discovery rate <0.5 and |log2 of fold change|>1) were screened, 123 of which were identified to be associated with the prognosis of GBM patients. The co-expression network of these prognosis-associated DEGs included 1405 interactions and 112 DEGs, and 6 functional modules were identified in the network. The prognostic prediction system was comprised of 63 signature genes with a specificity value of 0.929 and a sensitivity value of 0.948. GBM samples with good and bad prognosis in the TCGA, CGGA and GEO datasets were distinguishable by these signature genes (P=1.33×10(−6), 1.63×10(−4) and 0.00534, respectively). The co-expression network of signature genes with significant pathways was comprised of 56 genes and 361 interactions. Protein kinase Cγ (PRKCG), protein kinase Cβ (PRKCB) and calcium/calmodulin-dependent protein kinase IIα (CAMK2A) were important genes in the network, and based on the expression of these genes, it was possible to distinguish between samples with significantly different survival risks. In the present study, an effective prognostic prediction system for GBM patients was constructed and validated. PRKCG, PRKCB and CAMK2A may be potential prognostic factors for GBM.