Genomic instruments help to find Driver nodes in co-expression networks

Introduction & Objectives: One of the most important challenges in systems biology approaches has been finding driver nodes in network systems. The driver nodes are those who start the cascade of subsequent tiny changes to eventually reshape the whole system towards a new manner. These nodes can be used to engineer the system to yield significant and predictable impact and therefore are the best targets for drug design (1,2). Among different biological networks, co-expression networks have a special place as expression is almost always prior to other cellular changes. There are large-scale publicly available data sets from different tissues which are frequently used in network construction and analysis; however, focusing on the driver nodes is less studied in this field.Materials & Methods: Here we employed a Mendelian randomization (MR) approach to find genes whose expression changes the expression levels of a group of downstream genes. We applied the analysis on the largest expression data set generated to date (n ⁓32000) (3). We used the most significant single nucleotide polymorphism (SNP) associated with the expression level of each gene as instrumental variable (IV) and filtered out results with significant evidence of linkage confounding based on the heterogeneity test (HEIDI) in cis region of the gene (4). We tested each gene expression against the expression level of the nearby genes in cis region (1 Mb from each direction). We set the significance thresholds of both MR and HEIDI tests to Bonferroni corrected levels.Results: Our analysis returned 11162 expression quantitative trait genes (eQTGs) which cause changes in expression level of at least one downstream gene (p-value<2.61E-07 and HEIDI≥6.55E-07). Five eQTG hotspots, each affecting the expression level of more than 20 adjacent genes, are: E2F4, PRKAR2A, SLC12A4, RP11-96D1.5 and TRBV23-1. Driving nodes mainly consist of protein coding genes e.g. transcription factors (74%), regulatory RNAs e.g. antisense and lincRNAs (15%) and pseudogenes (7%).Conclusion: Our results help to find driver nodes in biological networks and hence to predict the resultant cellular and systemic phenotypes and also drug designs. As a next step, our study can be expanded to also include trans co-expression associations to find even more and robust eQTGs.

کلیدواژه ها:

Driver nodes ، network ، systems biology ، MR ، co-expression