Development of systematic methodology for the analysis of dynamic network model in systems biology

Abstract

Mathematical modeling of increasingly complex biological systems and datasets requires the process of analysis and calibration of a model base on experimental data, that is often challenging and a rate limiting step in model development. To facilitate this process, a systematic methodology was developed for (1) calibrating quantitative models of dynamic biological processes and (2) simulating cell-to-cell variability in a population. Utility of the developed methods were illustrated by a model of TRAIL (Tumor necrosis factor Related Apoptosis-Inducing Ligand)-induced cell death. First, a serial framework integrating analysis and calibration modules was proposed and various methods for global sensitivity analysis and global parameter estimation were compared. Adequacy of the network structure was checked by the analysis of global sensitivity to changes in concentrations of molecular species. The model could reproduce qualitative features of the system behavior observed in experiments or literature surveys. Then, rate parameters were ordered by the importance in the model using gradient-based and variance-based sensitivity indices, and the optimal number of parameters to be included in model calibration was systematically determined. Deterministic, stochastic and hybrid algorithms for global optimization were applied to estimate the values of the most important parameters by fitting them to time series data. The performance of these three optimization algorithms was tested for comparison. Secondly, cell-to-cell variability in a population was interpreted by computational modeling. Cells exhibit substantial phenotypic variation even though the cells are genetically identical. Variability due to intrinsic and extrinsic noises was simulated by three kinds of cell ensemble modeling. Furthermore, the origin of cell-to-cell variability in cell fate decision was identified by the schematic analysis over the whole network. Flux analysis and population modeling could...