Epigenetic programming: Genetic programming incorporating epigenetic learning through modification of histones
Introduction
“The major problem, I think, is chromatin [the dynamic complex of DNA and histone proteins that makes up chromosomes]. What determines whether a given piece of DNA along the chromosome is functioning, since it’s covered with the histones? What is happening at the level of methylation and epigenetics [chemical modification of the DNA that affects gene expression]? You can inherit something beyond the DNA sequence. That’s where the real excitement of genetics is now.”
J.D.Watson [23]
Until a few years ago, the role of histones (the family of proteins which DNA is wrapped around forming a super-coiled chromatin fiber) in molecular biology community was viewed as solely to help pack the long DNA into the tiny nucleus of eukaryotes’ cells. However, as the results of recent research suggest, the histones play a significant role in regulating the synthesis, repair, recombination and transcription of DNA [9], [19], [22]. For example, it is believed that regulation of the transcription mechanism of the same DNA via “histone code” during cell division controls the specialization of the cells yielding the well-known variety of cell types. Also, the gene expression regulated by “histone code” might control the variances in phenotypes (i.e. biochemistry, morphology, physiology and behavior) seen on different stages of life cycle of living organisms as developing, maturing and aging. Moreover, the onset of some genetically associated diseases (and even cancer) is viewed as a process triggered by both a sudden activation of the genes that “contribute” to the disease and/or deactivation of the genes that “fight” the disease. The histone code is regarded as an integrating link in the information pathway of epigenesis of living organisms. As illustrated in Fig. 1, the interaction between the phenotype and various environmental factors (such as food, viral infections, exposure to toxins, radiation, light and UV) leads to corresponding variations in the histone code, which in turn result in modified (beneficial or detrimental) gene expression.
Without touching the details of either the chromatin structure or the underlying chemical processed in histones, we would like to generalize the recently emerged findings that transcription of the genes in DNA is controlled by the surrounding chemical structure of histones. The acetylation of histones correlates with transcriptional activity of the corresponding DNA genes, while the methylation, with silencing (transcriptional inactivation) of genes.
In the proposed approach we extend the notion of inheritable genotype in genetic programming (GP) from the commonly considered model of DNA of the living organisms (the genetic program, typically represented as parse tree or corresponding linear representation as S-expression, prefix or postfix polish notation) into chromatin (DNA and histones). We attempt to mimic the naturally observed phenomenon of regulating gene expression via histone code into a software system featuring epigenesis, embedded into the simulated evolution (phylogenesis). Because we are mainly interested in beneficial modifications of histone code, which take place within the life cycle of evolved simulated organisms, we view the proposed approach of epigenetic programming as a form of epigenetic learning (EL) incorporated in GP. The objective of our work is to explore the effects of EL on the performance characteristics, and especially on the computational effort of evolution of autonomous agents in multi-agent systems (MAS).
Our work can be viewed as related to the methods of employing heuristics, phenotypic plasticity, Baldwin effect, inactive code [17], neutral code [5] and redundant representations [20] in GP. In contrast to the approaches of using heuristics [16], phenotypic plasticity [3] and Baldwin effect [6], the proposed learning mechanism does not imply direct manipulation of either the simulated DNA or the phenotype. Instead, the ontogenetic adaptation of phenotypes in our approach is achieved by controllable and inheritable gene expression mechanisms. The silenced genes can still comprise the genotype of the individual without affecting its phenotype. In addition to being biologically more plausible, such an approach offers (i) better phenotypic diversity of genotypically similar individuals and (ii) an efficient way to preserve the individuals from the destructive effects of crossover by explicit activation of the growing genetic combinations only when they are most likely to be expressed as beneficial phenotypic traits.
The previous work on the implications of emergent inactive code [5], [17], [20] on the performance of genetic programming has not been intended to investigate the effects of explicit manipulations of the inactive code, while the approach proposed in this paper centers on the computational efficiency of maintaining and explicitly manipulating the inactive code in a biologically plausible way.
The remainder of this document is organized as follows. Section 2 introduces the task, which we use to test our hypotheses – an instance of the general, well-defined yet difficult to solve predator–prey pursuit problem. The same section addresses the issue of developing the software architecture of the agents. Section 3 elaborates the main features of developed genetic programming, used to evolve the functionality of agents. The proposed mechanism of EL is introduced in Section 4. Section 5 presents empirically obtained results of the implications of EL on the performance of evolution. Conclusions are drawn in Section 6 where the anticipated directions for future research are mentioned.
Section snippets
Instance of predator prey pursuit problem
Currently, the main application areas of MAS are problem solving, simulation, collective robotics, software engineering, and construction of synthetic worlds [4]. Considering the latter application area and focusing on the autonomy of agents and the interactions that link them together [18], the following important issues can be raised: How can agents cooperate? What is the architecture they should feature so that they can achieve their goals? What approaches can be applied to automatically
Limiting the search space of genetic programming
We consider a set of stimulus–response rules as a natural way to model the reactive behavior of predator agents [8], which in general can be evolved using artificial neural networks, genetic algorithms, and genetic programming (GP). GP is a domain-independent problem solving approach in which a population of computer programs (individuals) is evolved to solve problems [10]. The simulated evolution in GP is based on the Darwinian principle of reproduction and survival of the fittest. In GP
Chromatin representation
In the developed approach of EL, incorporated in STGP, we consider the predator agents in STGP as simulated individuals passing through the phases of birth, ontogenetic adaptation and survival (reproduction) or death (Fig. 6).
At the phase of birth, the individual in STGP is represented as a single embryonic cell expressed by its respective chromatin. The ontogenetic adaptation is initiated by the simulated division of the embryonic cell into single germ cell and single somatic cell. Both cells
Values of parameters
The values of parameters of STGP used in our experiments are as follows: the population size is 400 genetic programs, the selection ratio is 0.1, including 0.01 elitism, and the mutation ratio is 0.02, equally divided between sub-tree mutation, transposition and histone modification. The termination criterion is defined as a disjunction of the following conditions: (i) fitness of the best genetic program in less than 300 and the amount of initial situations in which the prey is captured
Conclusion
We presented the results of our work inspired by recently discovered findings in molecular biology which suggest that histones play a significant role in regulating the gene expression in eukaryotes. Extending the notion of inheritable genotype in GP from commonly considered model of DNA to chromatin, we propose an approach of epigenetic programming as a way to incorporate the naturally observed phenomenon of regulated gene expression via modification of histones. Considering the individual as
Acknowledgements
The authors thank Katsunori Shimohara for his immense support of this research. The research was supported in part by the National Institute of Information and Communications Technology of Japan.
References (23)
- M. Benda, B. Jagannathan, R. Dodhiawala, On optimal cooperation of knowledge sources, Technical Report BCS-G2010-28,...
A robust layered control system for a mobile robot
IEEE Journal of Robotics and Automation
(1986)- A.I. Esparcia-Alcázar, K.C. Sharman, Phenotype plasticity in genetic programming: a comparison of darwinian and...
Multi-Agent Systems
(1999)Gene expression programming: a new adaptive algorithm for solving problems
Complex Systems
(2001)- et al.
Co-evolution and ontogenetic change in competing robots
- et al.
Evolving behavioral strategies in predators and prey
Emergence: From Chaos To Order
(1999)- et al.
Translating the Histone Code
Science
(2001) Genetic Programming: On the Programming of Computers by Means of Natural Selection
(1992)
Evolving Teamwork and Coordination with Genetic Programming
Cited by (18)
Epigenetic algorithms: A New way of building GAs based on epigenetics
2018, Information SciencesCitation Excerpt :This epigenetic approach is focused on the evolution of epigenetic agents to gain more knowledge about this field while we propose an algorithm and solve problems with it. The authors in [59] incorporate an explicitly controlled gene expression through histone modification in strongly-typed genetic programming (STGP) and call it, epigenetic programming. They propose a double cell representation of the simulated individuals represented by their respective chromatin structures.
Inference of compact nonlinear dynamic models by epigenetic local search
2016, Engineering Applications of Artificial IntelligenceCitation Excerpt :If a new gene is introduced via genetic mutation, that gene has the same probability of being active as the initial genes of the population (50%, in the current study). There has been some work to incorporate epigenetic learning into GP, notably by Tanev and Yuta (2008). In that case the focus was to model histone modification through a double cell representation as demonstrated in a predator-prey problem.