Metabolische Stimulus-Response-Experimente : Werkzeuge zur Modellierung, Simulation und Auswertung

Metabolic modeling has become a major activity in metabolic engineering in
recent years, in order to understand the complex regulation phenomena in a
living cell. In this thesis, a versatile tool has been developed to
support a modeler with the setup and refining of a model, that simulates
the data extracted from a rapidly sampled stimulus response experiment.

In the course of the modeling process, the modeler is typically not only
concerned with a single model but with sequences, alternatives and
structural variants of models. Supporting the modeling process of dynamic
biochemical networks based on sampled in vivo data requires more than just
simulation. For this purpose, the new concept of model families is
specified and implemented in this thesis. With this concept, a multitude
of similar models can be formulated in a single description by using
network and kinetic variants The concept allows to automatically navigate
in the space of models and to exclude biologically meaningless models on
the basis of elementary flux mode analysis.

An incremental usage of the measured data is supported by using splined
data instead of state variables. Powerful automatic methods are then
required to assist the modeler in the organization and evaluation of
alternative models. This tool has been developed as a computational
engine, intended to be built into a tool chain. By the use of automatic
code generation, automatic differentiation for sensitivity analysis, and
Grid computing technology, a high performance computing environment is
achieved. It supplies XML model specification and several software
interfaces.

The performance and usability of the tool of this thesis is illustrated by
several examples from ongoing research projects. An optimization algorithm
was developed, enabling the software tool to automatically carry through
the tasks of model variant switching and parameter fitting. The
computation results into a ranking of models variants fitting best to the
experimental data.