Processing signals with evolving artificial gene regulatory networks

Authors:

• Michał Joachimczak,
• Borys Wróbel

Abstract

Computational properties of gene regulatory networks (GRNs) are of great interest in the field of systems biology and, increasingly, in the field of artificial life. Understanding how GRNs work and evolve may help in elucidating the properties of real biological networks and in designing new biological networks for practical applications. Here we investigate the possibility to evolve artificial GRNs that can generate or process continuous signals represented by concentrations of artificial substances. We use a biologically-inspired model of regulatory networks. The way the nodes in the GRN (regulatory units) are connected and the weights of connections are encoded in a linear genome. A genetic algorithm is used to obtain GRNs that can solve problems with increasing difficulty. Some of these problems require performing simple mathematical operations and sustaining memory. We analyse if the solutions are general by presenting the GRNs with input patterns that were not used for fitness evaluation during evolution. We also briefly discuss the advantages of using biologically-inspired GRN-like systems for control problems and compare them with systems inspired by neural networks.