Research Of Intelligent Algorithms For Control And Optimization In Metabolic Pathways - Parameter Estimation

Systems Biology is a new field in biology that aim at understanding biological systems at systems level. With the rapid progress in molecular biology, furthered by technologies for making comprehensive measurements on DNA sequence, gene expression profiles, protein-protein interactions, serious attempts to understand biological systems as systems are now almost feasible. Handling this high-throughput experimental data places major demands on computer science, including database processing, modeling, simulation, and analysis.The parameter estimation (inverse problem) of nonlinear dynamic biochemical pathways which has been stated as a nonlinear programming problem subject to nonlinear differential-algebraic constrains has been discussed. The problem is frequently ill-conditioned and multimodal, traditional local optimization (gradient-based) methods fail to arrive at satisfactory solutions. Intelligent algorithms had been widely used in nonlinear programming problems for the efficiency, convergency and robustness. The purpose is to find an intelligent algorithm which suitable for the parameter estimation of the dynamic biochemical systems. The main contents of this dissertation are as follows:First, we introduce the current state on intelligent optimization algorithms and also do some theoretical researches on the algorithms. An improved Quantum-behaved Particle Swarm Optimization is proposed for the local convergence of the algorithm. The improved QPSO employs a mutation operation exerted on the global best position to enhance the search ability of the QPSO algorithm. The improved algorithm has a better performance as showed by the comparative experiments.Second, a case study considering the estimation of 36 parameters of a nonlinear biochemical dynamic model has been taken as a benchmark. Several intelligent optimization algorithms have been explored to the problem. We use a weighted distance measure between experimental and predicted values of the 8 state variables as the objective function. The model for solving the inverse problem has been created. Experiments show that quantum-behaved particle swarm optimization algorithm can estimate the 36 parameters with high accuracy and successfully reconstruct the metabolic networks.Third, the analysis and reconstruction of genetic regulatory network based on QPSO algorithm have been discussed. The problem based on the S-system model is defined as an estimation problem of the S-system parameter. The sum of squared errors between experimental values and predicted values is used as the objective optimization function in the basic method, but it could predict only a very small number of parameters and the convergence rate is low. The gradual optimization strategy based on QPSO algorithm is proposed with the sparsity of the genetic network. The optimization procedure became simple with the futile parameters gradually fixed. The dynamic of a small genetic network constructed with 60 parameters for 5 networks variables is successfully inferred in experiments by the improved algorithm.