Kronfeld, Marcel and Dräger, Andreas and Aschoff, Moritz and Zell, Andreas

On the Benefits of Multimodal Optimization for Metabolic Network Modeling

German Conference on Bioinformatics (GCB 2009), Halle (Saale), Germany, Lecture Notes in Informatics vol. P-157, German Informatics Society, 2009, pp. 191-200


Abstract

The calibration of complex models of biological systems requires numerical simulation and optimization procedures to infer undetermined parameters and fit measured data. The optimization step typically employs heuristic global optimization algorithms, but due to measurement noise and the many degrees of freedom, it is not guaranteed that the identified single optimum is also the most meaningful parameter set. Multimodal optimization allows for identifying multiple optima in parallel. We consider high-dimensional benchmark functions and a realistic metabolic network model from systems biology to compare evolutionary and swarm-based multimodal methods. We show that an extended swarm based niching algorithm is able to find a considerable set of solutions in parallel, which have significantly more explanatory power. As an outline of the information gain, the variations in the set of high-quality solutions are contrasted to a state-of-the-art global sensitivity analysis.


Downloads and Links

[pdf] [pdf]


BibTeX

@inproceedings{Kron09NichingGCB,
  author = {Kronfeld, Marcel and Dr\"ager, Andreas and Aschoff, Moritz and Zell,
	Andreas},
  title = {{On the Benefits of Multimodal Optimization for Metabolic Network
	Modeling}},
  booktitle = {German Conference on Bioinformatics (GCB 2009)},
  year = {2009},
  editor = {Grosse, Ivo and Neumann, Steffen and Posch, Stefan and Schreiber,
	Falk and Stadler, Peter},
  volume = {P-157},
  number = {978-3-88579-251-2},
  series = {Lecture Notes in Informatics},
  pages = {191--200},
  address = {Halle (Saale), Germany},
  month = sep,
  publisher = {German Informatics Society},
  abstract = {The calibration of complex models of biological systems requires numerical
	simulation and optimization procedures to infer undetermined parameters
	and fit measured data. The optimization step typically employs heuristic
	global optimization algorithms, but due to measurement noise and
	the many degrees of freedom, it is not guaranteed that the identified
	single optimum is also the most meaningful parameter set. Multimodal
	optimization allows for identifying multiple optima in parallel.
	We consider high-dimensional benchmark functions and a realistic
	metabolic network model from systems biology to compare evolutionary
	and swarm-based multimodal methods. We show that an extended swarm
	based niching algorithm is able to find a considerable set of solutions
	in parallel, which have significantly more explanatory power. As
	an outline of the information gain, the variations in the set of
	high-quality solutions are contrasted to a state-of-the-art global
	sensitivity analysis.},
  pdf = {http://www.cogsys.cs.uni-tuebingen.de/publikationen/2009/Kron09NichingFinal.pdf},
  url = {http://www.gcb2009.de/program.php}
}