Information propagation in the long-term behavior of gene regulatory networks

Abstract

Gene regulatory networks (GRNs) constantly receive, process and send information. While stochastic in nature, GRNs respond differently to different inputs and similarly to identical inputs. Since cell types stably remain in restricted subsets of the possible states of the GRN, they are likely the dynamical attractors of the GRN. These attractors differ in which genes are active and in the amount of information propagating within the network. Using mutual information (I) as a measure of information propagation between genes in a GRN, modeled as finite-sized Random Boolean Networks (RBN), we study how the dynamical regime of the GRN affects I within attractors (IA). The spectra of IA of individual RBNs are found to be scattered and diverse, and distributions of IA of ensembles are non-trivial and change shape with mean connectivity. Mean and diversity of IA values maximize in the chaotic near-critical regime, whereas ordered near-critical networks are the best at retaining the distinctiveness of each attractor's IA with noise. The results suggest that selection likely favors near-critical GRNs as these both maximize mean and diversity of IA, and are the most robust to noise. We find similar IA distributions in delayed stochastic models of GRNs. For a particular stochastic GRN, we show that both mean and variance of IA have local maxima as its connectivity and noise levels are varied, suggesting that the conclusions for the Boolean network models may be generalizable to more realistic models of GRNs. © Springer-Verlag Berlin Heidelberg 2011.