MODEL IDENTIFICATION IN THE BIOCHEMICAL SYSTEMS THEORY

Abstract

Mathematical modeling of biological systems using the power-law formalism, the Biochemical Systems Theory (BST) coupled with high-throughput biological measurements transform the model identification into an inverse problem of estimating model parameters from experimental data. Despite the large number of publications on this topic, this task remains the bottleneck in the application of BST modeling in biologically related area. However, many challenges arise from the same underlying problem; incomplete and noisy measurements lack the necessary information in order to accurately estimate the model parameters. This is a parameter identifiability problem. Thus, the focus of my PhD work is parameter-identifiability-centric modeling of biochemical networks using the BST models. The applications of the methods developed for identifiability analysis to two inverse modeling problems within the BST point to the lack of parametric identifiability as the root cause of the difficulty faced in the inverse modeling. Motivated by the results of the analysis to S-systems and GMA models, we developed methods based on nonlinear regression to test the identifiability of decoupled and lin-log systems. The next work deals with design of experiments (DOE) to generate information-rich data that improves parameter identifiability. To this end a multiobjective optimization design criterion that accounts for curvature effects was developed. The final goal of my PhD project was to integrate the tasks of identifiability analysis, DOE and parameter estimation together into a useful MATLAB tool.