Metanetworks of artificially evolved regulatory networks

TL;DR

This paper investigates the structure and robustness of metanetworks formed by evolved Boolean networks in genotype and phenotype spaces, revealing giant clusters, hierarchical organization, and high robustness in evolved populations.

Contribution

It introduces a model linking genotype mutations to phenotype attractors, analyzing the resulting metanetwork structures and their properties in evolved versus random populations.

Findings

Evolved populations form giant clusters in genotype space.

Metanetworks exhibit hierarchical $ppa$-core structures.

Evolved networks are highly robust under node and connection removal.

Abstract

We study metanetworks arising in genotype and phenotype spaces, in the context of a model population of Boolean graphs evolved under selection for short dynamical attractors. We define the adjacency matrix of a graph as its genotype, which gets mutated in the course of evolution, while its phenotype is its set of dynamical attractors. Metanetworks in the genotype and phenotype spaces are formed, respectively, by genetic proximity and by phenotypic similarity, the latter weighted by the sizes of the basins of attraction of the shared attractors. We find that populations of evolved networks form giant clusters in genotype space, have Poissonian degree distributions but exhibit hierarchically organized $\kappa$-core decompositions. Nevertheless, at large scales, they form tree-like expander graphs. Random populations of Boolean graphs are typically so far removed from each other genetically that they cannot form a metanetwork. In phenotype space, the metanetworks of evolved populations are super robust both under the elimination of weak connections and random removal of nodes.