Spieth, Christian and Streichert, Felix and Supper, Jochen and Speer, Nora and Zell, Andreas

Feedback Memetic Algorithms for Modelling Gene Regulatory Networks

IEEE Symposium on Computational Intelligence and Computational Biology, San Diego, USA, IEEE Press, 2005, pp. 61-67


Abstract

In this paper we address the problem of finding gene regulatory networks from experimental DNA microarray data. We focus on the evaluation of the performance of memetic algorithms on the inference problem. These algorithms are used to evolve an underlying quantitative mathematical model. The dynamics of the regulatory system are modeled with two commonly used approaches, namely linear weight matrices and S-systems. Due to the complexity of the inference problem, some researchers suggested evolutionary algorithms for this purpose. We introduce memetic enhancements to this optimization process to infer the parameters of sparsely connected nonlinear systems from the observed data. Due to the limited number of available data, the inferring problem is underdetermined and ambiguous. Further on, the problem often is multimodal and therefore appropriate optimization strategies become necessary. We propose a memetic method, which separates the overall inference problem into two subproblems to find the correct network: first, the search for a valid topology, and secondly, the optimization of the parameters of the mathematical model. The performance and the properties of the proposed methods are evaluated and compared to standard algorithms found in the literature.


Downloads and Links

[pdf]


BibTeX

@inproceedings{2005_20,
  author = {Spieth, Christian and Streichert, Felix and Supper, Jochen and Speer,
	Nora and Zell, Andreas},
  title = {Feedback Memetic Algorithms for Modelling Gene Regulatory Networks},
  booktitle = {IEEE Symposium on Computational Intelligence and Computational Biology},
  year = {2005},
  pages = {61-67},
  address = {San Diego, USA},
  month = nov,
  publisher = {IEEE Press},
  abstract = {In this paper we address the problem of finding gene regulatory networks
	from experimental DNA microarray data. We focus on the evaluation
	of the performance of memetic algorithms on the inference problem.
	These algorithms are used to evolve an underlying quantitative mathematical
	model. The dynamics of the regulatory system are modeled with two
	commonly used approaches, namely linear weight matrices and S-systems.
	Due to the complexity of the inference problem, some researchers
	suggested evolutionary algorithms for this purpose. We introduce
	memetic enhancements to this optimization process to infer the parameters
	of sparsely connected nonlinear systems from the observed data. Due
	to the limited number of available data, the inferring problem is
	underdetermined and ambiguous. Further on, the problem often is multimodal
	and therefore appropriate optimization strategies become necessary.
	We propose a memetic method, which separates the overall inference
	problem into two subproblems to find the correct network: first,
	the search for a valid topology, and secondly, the optimization of
	the parameters of the mathematical model. The performance and the
	properties of the proposed methods are evaluated and compared to
	standard algorithms found in the literature.},
  isbn = {0-7803-9387-2},
  url = {http://www.cogsys.cs.uni-tuebingen.de/publikationen/2005/spieth05feedback.pdf}
}