The artificial epigenetic network

Abstract

The term epigenetics refers to typically heritable biological mechanisms which facilitate stable yet reversible modifications of gene expression or phenotype state, without alteration of the underlying genetic code. More specifically, epigenetic mechanisms allow organisms to control which genes are active at a given time. In eukaryotes, epigenetic mechanisms have essential roles in gene regulation, cellular differentiation and genetic packaging. These epigenetic mechanisms give rise to functionality which DNA alone is generally incapable of providing.This thesis takes inspiration from the fields of genetics and epigenetics, and builds a computational model which captures the beneficial properties of epigenetics in silico. This computational model is referred to as the artificial epigenetic network. The artificial epigenetic network can dynamically control which genes within the network are active at a given time, allowing certain groups of genes to become specialised towards specific aspects of a task.Hence, the artificial epigenetic network can contain many different regulatory circuits, each with specific properties. This gives the networks the ability to more readily express a wider range of dynamical behaviours, which were found to produce a number computational benefits. The artificial epigenetic network is applied to a diverse range of control tasks, each with varying dynamics, to ascertain how the functionality of the artificial epigenetic structures effects the functionality of the network. An emergent property is that the epigenetic structures can partition the network into functional units corresponding to the logical decomposition of the tasks, and control these units with a switch like behaviour. This provides an interface, where a user can gain control over the complex dynamics of the target domain via the activation or deactivation of these switches.3

Artificial Epigenetic Networks: Automatic Decomposition of Dynamical Control Tasks Using Topological Self-Modification

“…Self-modification was also explored in [31] within the context of a model of mobile DNA applied to Boolean networks, in which the author found it to be beneficial in terms of access and stability of attractors. We have also explored a simpler model of chromatin remodeling in which subtasks are prespecified [8], and published initial results using the current approach [9] which have been significantly extended in this paper.…”

“…For more information, see http://creativecommons.org/licenses/by/3.0/ tasks. This paper builds upon an earlier model [8], in which task decomposition was prespecified, and upon initial results reported in [9]. This paper is structured as follows.…”

Biochemical connectionism

“…In [105] and [106], we considered the benefits of introducing an analogue of chromatin remodelling to a connectionist architecture. This works by introducing extra nodes into the network, each of which is used to regulate the activity of a chromatin frame.…”

“…Starting with the pendulums hanging beneath the carts (b), the objective is to swing the pendulums to an upright position (e) and then maintain them in this position for a defined period of time. Adapted from [105].…”

“…Coupling between the carts makes this problem challenging [38], since collisions can readily propagate and destabilise the system. However, using an analogue of chromatin remodelling within a sigmoidal recurrent network, we were able to evolve controllers that could control five inverted pendulums at once, a behaviour that could not be achieved readily with the non-reconfigurable networks alone [105]. Analysis of the evolved networks suggests they achieve this by modulating a relatively small number of nodes.…”