Gene Regulation in a Particle Metabolome Simon Hickinbotham, Edward Clark, Susan Stepney, Tim Clarke and Peter Young Abstract— The bacterial genome is well understood by biol- ogists. Although its efficiency and adaptability should make it a good model for evolutionary algorithms, the bacterial genome is tightly coupled with the components of the bac- terial metabolism, referred to here as the metabolome. This paper explores an approach to modelling an artificial bacterial metabolome in an efficient and modular manner, so that analogues of bacterial genome organisation and gene regulation can be implemented in evolutionary algorithms. We propose a particulate model of bacterial metabolic pathways in which the constituents drift in a fixed, limited space and obey a limited set of biologically plausible reaction rules. The potential of this model is demonstrated by creating a network that is capable of appropriate behavioural switching that can be observed in bacteria. I. I NTRODUCTION We observe from the bio-diversity present in nature the power of biochemical evolution to build. This property is not so clearly demonstrated in artificial evolutionary algorithms (EAs), which tend to serve more as optimisers [1]. Current attempts to improve the ability of EAs to create functional structures take one of three routes: modifying canonical EA mechanisms via (for example) new configurations of crossover or mutation [2], augmenting the system with other computational devices such as neural networks [3], or taking further inspiration from biology [4]. Our ambition is to take the biological route, but to take a few steps back from current artificial EA systems and draw inspiration directly from so-called “simple” unicellular life forms, specifically the prokaryotes (bacteria) and their postulated precursors in the evolution of the early earth. Central to this approach is the idea that the metabolism acts as a sorting-house for signals from the environment and from the genome. The metabolism is attractive for several reasons: it reflects the response of the genome to the current state of the environment on a range of time scales; it is able to deal with information in a variety of forms; it is capable of switching behaviour as a result of change in stimulus. These desirable properties are all coded for in the bacterial genome. Given that bacteria demonstrate that a genome can be used to build information-processing “factories”, why have they not been used more extensively as templates for evolutionary algorithms? The answer is that the bacterial metabolism is a highly complex network of interacting three dimensional structures which biologists have been working on mod- elling and understanding for the past 150 years [5]. These This work is part of the Plazzmid project, funded by EPSRC grant EP/F031033/1. All authors are affiliated with the York Centre for Com- plex Systems Analysis, University of York, Heslington, York YO1 5DD, UK. Simon Hickinbotham; email: sjh@cs.york.ac.uk), Ed Clark and Susan Stepney are at the Department of Computer Science, Tim Clarke is at the Department of Electronics and Peter Young is at the Department of Biology. works all have the goal of understanding elements of the prokaryote metabolism. Our goal is different: developing a model metabolism that is sufficiently rich to allow useful experiments with regulation of gene expression within EAs to be carried out. We find the term metabolome useful here. The metabolome is simply the set of (small) molecules that make up a biological system along with reactions that occur between them, whereas the metoblism defines the “physics” of the system, which is not encoded on the genome. We require an appropriate abstraction of bacterial metabolism that preserves their complexity and robustness but which can also be encoded in some artificial genetic representation such that evolutionary experiments can be conducted. We want to see what computational features and problems exist at the metabolic level, so that we don’t waste time constructing genetic systems that do not encode them with appropriate detail. In this sense, the model is “top down”. But we want to emphasise that we are aiming for a “pluggable” model, in which the representation of the metabolic processing unit is separated from the genomic and protein/enzyme/molecular representations. Different applica- tions of this model will require different resolutions in these three domains. We do not therefore concern ourselves with specific issues of three dimensional shape of metabolites, with all the implications for protein folding and binding that follow. Nor do we want to model our metabolism as a continuous distribution of concentrations of solvents, since that removes the possibility for local variation of individual metabolites that is a necessary part of evolution. Finally, we are not concerned about faithfully simulating biological re- action rates and metabolite counts of bacterial metabolomes, since we recognise that the computational burden would probably be too heavy even for that. This paper describes our particle metabolome model, and demonstrates how it can be used to engineer self-regulatory control in a virtual organism. As a test of the versatility of the resulting metabolome, we describe how it can be used to model gene regulatory control of the enzyme complement of an artificial metabolism. We take inspiration from diauxy, the regulation of the metabolism of lactose, which is an alternative and less energetically beneficial dietary substrate to glucose [6]. Note that our goal is to demonstrate that this type of control can be implemented in the system we describe. II. THE PARTICLE METABOLISM We are constrained by the idea that metabolome particles can not make reference to some cell-level instruction set that determines what metabolic reactions are permissible, since our long term goal is to evolve metabolites that are