Development of System Identification Technique Based on Real-Coded Genetic Algorithm

Recent advances of powerful new technologies such as DNA microarrays provide a mass of gene expression data on a genomic scale. One of the most important projects in post-genome-era is the system identification of gene networks by using these observed data. We previously introduced an efficient numerical optimization technique by using time-course data of system components, which is based on real-coded genetic algorithm (RCGAs) to estimate the reaction coefficients among system components of a dynamic network model called S-system [3] that is a type of power-low formalism and is suitable for description of organizationally complex systems such as gene expression networks and metabolic pathways. This technique with the combination of one of the crossover operators for RCGAs called unimodal normal distribution crossover (UNDX) [1] with the alternation of generation model called minimal generation gap (MGG) [2] showed remarkable superiority to the simple GA in case of simple oscillatory system [4]. However this case study belongs to a comparative easy inverse problem; the number of system components was 2 and the estimated parameters was 12. For application to gene networks including huge number of estimated parameters, our new optimization techniques have to be adapted to inverse problem with more strict circumstances. In this paper, we shall attempt to the inference of the interactions in more large scale of gene expression networks. In the case study, we also propose new efficient approaches to narrow down the candidates that explain the observed time-courses within the immense huge searching space of parameter values.