Structure Identification of Biochemical Systems with Genetic Algorithms

Modern methods of genomics and proteomics are beginning to yield data of a quantity and quality unimaginable just a decade ago. Already, sophisticated nuclear magnetic resonance and mass spectrometry techniques are able to quantify hundreds and sometimes thousands of simultaneously measured metabolites. These metabolic profiles are usually obtained as snapshots, but could be generated as dense time series that reveal the dynamic trends in many or all metabolites of an organism, following some stimulus. Metabolic profiles contain enormous information about the flux distribution, and also the regulation, of metabolic pathways in vivo. This information is not immediately explicit though, but requires adequate analytical and computational methods of retrieval and interpretation. We propose to address this information retrieval with a novel approach that combines an established modeling framework for metabolic systems, Biochemical Systems Theory (BST), with computational methods of parameter estimation that are based on a Genetic Algorithm (GA). The key feature of this merging of techniques is that parameter values in biochemical systems models are directly and uniquely related to both the flow structure and the regulation of the modeled network, which allows interpretation of estimated parameter values in terms of possible and probable network structures. BST, and particularly its implementation in the form of S-system differential equations, provides powerful tools for the modeling and analysis of complex metabolic networks. Of crucial benefit in the present context is that every S-system model has the same homogeneous mathematical structure, namely,