| Summary: | 碩士 === 國立中正大學 === 化學工程研究所 === 92 === We have developed three different global/local optimization methods for estimating model parameters of genetic regulatory networks. there are two major challenges in the estimation of parameter values from different time series data. First, most nonlinear regressions belong to gradient-based nonlinear optimization methods so that the solution quality strongly depends on the provided initial guess. Moreover, such gradient-based methods may yield a local minimum. In this study, we introduce different global optimizations to avoid this problem. The estimates are then provided for a local minimization method to yield the more accuracy result. The second challenge is the optimization algorithm, whatever a local or global method, must solve the system of differential equations during every iteration. Numerical integration failure due to the stiffness is the major problem in the estimation. we applied a improved modified collocation method to avoid numerical integration. The improved modified collocation method can be directly employed to the dynamic profile not only using smooth data but also those including 5% noise data to the estimation problem. In this representation, we successfully determine the parameters and structure of two genetic regulatory models.
|
|---|
