Behavioural Selection Pressure Generates Hierarchical Genetic Regulatory Networks

The eld of 'evo-devo'|evolutionary developmental biology|is making rapid inroads to biological questions that encompass phylogenetic evolution and ontogenetic development, speci cally in regards to genetic regulatory networks (GRNs). However, there is relatively little understanding so far of how selection pressure shapes GRNs (Carroll 2000). We have shown that by enhancing evolutionary algorithms with genetic regulatory networks, it is possible to not only evolve simulated agents that can perform behavioural tasks, but it is also possible to analyze both evolved GRNs, and the evolutionary history of them in the evolving population (Bongard 2002). Here we show that successful evolutionary runs produce heirarchical GRNs: there is a dominant unidirectional ow in gene regulation, and relatively few cyclical gene regulation pathways. Arti cial Ontogeny extends the genetic algorithm to include ontogenetic development. In the results presented below, agents are tested for how fast they can travel over an in nite horizontal plane during a pre-speci ed time interval. The tness determination is a two-stage process: the agent is rst grown from a GRN (the growth phase), and then evaluated in its virtual environment (the evaluation phase). See (Bongard 2002) for methodological details.