Large scale model identification in systems biology

A An important consequence of the Human Genome Project is that scientists are challenged to move towards a new view of biology---a systems level approach. Current molecular biology is producing high-throughput quantitative data as a result of advances in experimental techniques. This has led to a need to integrate the experimental research with computational tools for a better understanding of the complex biological systems. With the wide use of biological processes in the industry there is also a need to develop sophisticated computational tools for the design, optimization and control of the process for optimal performance.; In this work, an approach is proposed for model identification of biological networks. The scheme has general application to modeling a wide range of cellular processes, which include gene regulatory networks, signaling networks and metabolic networks. The proposed scheme includes a State Regulator Problem (SRP) that provides estimates of all system unknowns using the available systems knowledge and limited experimental data. The full set of estimates is used for parameter estimation to develop large scale dynamic models for the biological network. A framework is developed for model refinement with the availability of new systems knowledge and experimental data. The framework utilizes the Fisher Information Matrix (FIM) theory for determining the optimal experiment design and measurement set to generate information rich experimental data for improving the model. The model refinement is performed in an iterative fashion until an acceptable model is obtained. The application of the framework is demonstrated on signaling networks of cell apoptosis and the MAP kinase cascade.; An application of large scale models is demonstrated for metabolic engineering. A bilevel optimization framework is developed to determine optimal genetic alterations for the manipulated reactions to maximize desired cellular behavior. It is demonstrated that it is suboptimal from the standpoint of productivity to induce the genetic alterations at the start of the production process. The application of the bilevel framework is demonstrated for glycerol and ethanol production in batch fermentation of Escherichia coli. The framework is integrated with sophisticated control algorithms for optimal control both at the genetic and bioreactor level. Practical issues, such as plant-model mismatch, measurement error and unmeasured disturbances are addressed. The application to a bioreactor scheduling problem is also discussed.