Study of signaling and regulatory networks by computational methods

Abstract

In order to communicate with extracellular and intracellular environments, cells have to be able to receive signals and via signaling pathways pass them to nucleus, where the genome responds through cascades of gene regulatory networks (GRNs). Many signal receptors on the cell membranes consist of extracellular part that can bind particular molecules, and intracellular part that accomplishes a conformation change or open binding sites for other proteins after the extracellular binding. Consequently, some proteins/enzymes are activated or recruited, and the signaling transduction then onsets via protein-protein interactions (PPIs). Once the signal reaches nucleus and related transcription factors (TFs) are activated, they will in turn regulate the gene transcription and change the gene expression. However, systematic identification of signaling network components and construction of GRNs are not easy to be achieved by pure experiments, which are costly and time-consuming.

In this dissertation, we construct a workflow to study the signaling and regulatory networks by applying computational methods on public PPI and omics data. We first explore the phenotypic profiles of genes using a cross-species phenotype network and compile the them into a gene-phenotype association database; then a webserver is proposed using omics data to construct GRNs; next the phenotypic profiles, together with domain-domain interactions, phylogenetic profiles and expression profiles, are used as features to refine the PPI network using random forest classifier. The random walk with restart algorithm is applied to the refined PPI network to identify the signaling network components. The GRNs centered by the predicted TFs are finally constructed and with related omics data or putative binding information. With this workflow, we have successfully identified important TFs, including Pou5f1, Sox2 and Nanog, involved in the cell reprogramming mediated by vitamin C. For interaction partners and co-regulatory targets of core pluripotency factors, large parts of the results (154 out of 229 interactions) have been confirmed by literature. This workflow can also be applied to decode the signaling and regulatory networks via which the genome responds to the environmental stimuli (i.e. drug treatment). The results will not only help biologists and clinicians to better understand biological processes and diseases systematically, but also lead to further studies, such as drug repurposing.