e23162 Background: Cancer is characterized by a variety of heterogeneous genomic and transcriptomic patterns involving highly complex signaling biological pathways. The problem of identification of the factors driving tumor progression becomes even more challenging due to intricate interaction mechanisms between these pathways. Using novel approaches in machine learning, we demonstrate the ability to quantitatively describe characteristic signaling patterns in cancer based on transcriptomic data Methods: We used RNASeq data from 20531 genes in 174 samples of GBM from The Cancer Genome Atlas including 5 major histological subtypes – Classical, G-CIMP, Mesenchymal, Neural, and Proneural, anddeveloped predictive computational framework for molecular subtype differentiation from normal tissue relying on variance based gene selection and random forest algorithm. Results: We obtained a few key findings – (1) genes from cell signaling pathways alone differentiate each subtype from normal tissue with 100% accuracy; (2) predictive genes are specific to each subtype; (3) inferred pathway interactions are also specific to each subtype; (4) typically most of the predictive genes involved in signaling are down-regulated in tumor compared to normal tissue (MAPT, PRKCG, PDE2A, RYR2, ATP1B1, GRN1, GNAO1), however, in each subtype we observed a smaller subset of predictive genes which are highly up-regulated in tumor (ID3, FN1, JAG1, F2R, COL4A1, EDAR, CDK2, CDK4, MFNG, BIRC5, CCNB2). We detected and quantitatively evaluated characteristic signaling pathway involvement across the GBM subtypes for MAPK, RAP1, RAS, Notch, PI3K-Akt, mTOR, FoxO, Jak-STAT, Wnt, cAMP, and Calcium Signaling, providing a unique approximation for each subtype signaling profile. Conclusions: In this study, we identified gene expression profiles and associated signaling pathways for distinguishing GBM Multiforme subtypes from normal tissue. We observed and described a dense complex picture of interacting signaling pathways. The detected interactions may provide clinical insights and could be used to identify potential therapeutic targets, however, more research is needed to confirm this.
Integrative analysis of gene expression profiles reveals specific signaling pathways associated with pancreatic duct adenocarcinoma