Maximizing local information transfer in Boolean networks

Abstract

We study a Boolean network model such that rules governing the time evolution of states are not given a priori but emerge from the maximization process of local information transfer and are stabilized if possible. We mathematically derive the class of rules that can be stabilized. With the presence of small noise, those stabilized are such that their output depends on a unique input. We confirm the prediction of the theory by numerical simulation. We argue that the stabilized rules have generic properties of real-world gene regulatory networks, being both critical and highly canalized.

Keywords

• mutual information
• Bayesian inference
• gene regulatory networks
• canalization
• criticality