Prognostic risk model construction and molecular marker identification in glioblastoma multiforme based on mRNA/microRNA/long non-coding RNA analysis using random survival forest method.

We aim to identify novel molecular signatures for prognosis prediction of glioblastoma multiforme (GBM). The expression and microarray data of GBM were downloaded from The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO). Differentially expressed mRNAs, microRNAs (miRNAs) and long non-coding RNAs (lncRNAs) between GBM and normal samples were identified by differential expression analysis using Bayesian T-test. Functional enrichment analysis was performed to identify GBM associated functions and pathways. A subset of signature mRNAs was selected from differentially expressed mRNAs and used to build a risk model for GBM using random survival forest (RSF) method. The performance of the model in prognosis prediction was validated using an independent validation dataset. A competing endogenous RNA (ceRNA) network was then constructed and key prognostic markers were identified from the network by survival analysis. In total, 905 mRNAs, 24 miRNAs and 403 lncRNAs were identified to be differentially expressed between GBM and normal samples. Functional and pathway items such as p53 signaling and PI3K/Akt signaling were significantly enriched by differentially expressed mRNAs. The RSF risk model showed a high performance in prognosis prediction for both training and validation dataset. The ceRNA network provided a comprehensive view of the interplays between differentially expressed mRNAs, miRNAs and lncRNAs. Among the ceRNA network, p21 (RAC1) activated kinase 1 (PAK1) and synaptic vesicle glycoprotein 2B (SV2B) were identified as key prognosis associated markers. The RSF risk model and key prognostic markers may contribute to GBM diagnosis in future clinical practice.