Global Optimization for Parameter Estimation and Yield Improvement in Complex Metabolic Pathway Systems
by
Edward P Gatzke
Department of Chemical Engineering, University of South Carolina
Coauthors: Pradeep K. Polisetty, Ebeerhard O. Voit

Biological systems can often be modeled as complex reactive networks involving large numbers of interacting species. The complexity of these networks ranges from relatively simple metabolic pathways to complex cell-signaling systems. Recent instrumentation improvements have allowed for increased experimental data collection. The dynamic response of biological systems to environmental stimuli can now be captured for modeling and analysis purposes. Researchers currently are working to create dynamic models of complex networked systems from experimental data, improve and modify complex systems, and control the dynamic response of the system when possible. Process systems engineering methods for modeling, design, and control can be used find solutions to these biological problems. In many instances, problems in this area can be formulated as numerical optimization problems. The resulting mathematical programming problem is often nonconvex, either due to discrete decision variables and / or algebraic nonlinearity in the constraints or objective function. Deterministic methods for solution of nonconvex optimization problems must be considered in order to provide guarantees on the quality of the resulting solution, as well as rigorous bounds on intermediate solutions. Advances in computational hardware have made possible large-scale parallel solution methods, allowing larger problems to become tractable and moderate problems to be solved in real-time. A parallel bounds contraction method for rapid solution of nonconvex nonlinear programming problems will be presented. Additionally, a decomposition method for the deterministic solution of mixed-integer nonlinear programming problems involving factorable, nonseparable, nonconvex constraints will be presented. Both methods are based on formulation of a nonconvex lower bounding problem using piecewise linear outer approximations of the original convex function relaxations. Results will be presented from applications areas including network modeling / nonlinear parameter estimation using time-series data, yield optimization considering optimal network modification, and unit level prioritized objective feedback control.

Date received: February 27, 2007