Modeling based process development of fed-batch bioprocesses: L-valine production by Corynebacterium glutamicum

A framework for modeling and optimization was written in the Matlab environment. The system allows the use and simulation of dynamic multivariable experiments. Models can be fitted to the experimental data and the accuracy of the parameter estimates can be regarded. Also the accuracy of the model fit can be addressed and the different models can be compared. Importantly, the programs also provide options for the design of informative experiments. These can be either experiments for discrimination among rival models, for improvement of the parameter estimation or for improvement of the production process with respect to, for example, the overall volumetric productivity. Constraints can be provided on the designed experiments. The constraints which were used in this work, included limits on the allowed feedrates and initial conditions as well as limits on the expected concentrations and suspension volume throughout the designed experiment.A general sequential experimental design procedure was presented where first model discriminating experiments are suggested until one model structure would be chosen. If necessary, subsequent experiments can then be designed for improvement of the parameter estimation. For the design of experiments for parameter estimation, a D-optimal design criterion was implemented. Five different model discriminating design criteria were adapted from criteria found in literature, for the use with dynamic multivariable experiments and compared in a simulative study with two examples. The one criterion that used the system entropy lead to more insecure results. The positive influence of the use of a model variance, which was used in three of the criteria, was not clear in the current examples, but the model variance accounted for about 90% of the calculation time. With larger models and longer experiments, this is bound to get even more.The system for modeling and experimental design was used for process development of L-valine production by a genetically modified C. glutamicum. With four experiments, a model was developed and selected. Two different model discriminating criteria were used for the design of these experiments. The selected model was used for calculation of optimal feed trajectories for optimization of the total volumetric productivity of the valine production process. This process was performed without further control strategies, leading to the production of 6.2 mMh-1 L-valine in a fermentation of 30 h, an 11% increase compared to the best prior experiments. The model was shown to predict especially the glucose uptake rate very satisfactory in the optimized experiment, i.e. in the region of interest. Also some discrepancies between the model predictions and the actual experimental results were identified, providing indications for model improvement. These discrepancies are caused by the strong extrapolation of the model outside the conditions, which were used before. Therefore, the use of optimized feed trajectories is also recommended during model development. Furthermore, interesting information on the used biological system under process relevant conditions was gathered, providing important indications for further research and possibly also strain improvement.

Grantor dc:publisher

Publikationsserver der RWTH Aachen University

Year dc:date

2005

Author dc:creator

• Brik Ternbach, Michael Alexander

Contributors dc:contributor

• Büchs, Jochen

• info:eu-repo/classification/ddc/660
• Technische Chemie
• Bioverfahrenstechnik
• Prozessentwicklung
• Modellgetriebene Entwicklung
• Modellierung
• Modellwahl
• Parameteridentifikation
• Corynebacterium glutamicum
• Fed-Batch

Statement dc:rights

• info:eu-repo/semantics/openAccess

Language dc:language

eng

Identifier

https://publications.rwth-aachen.de/search?p=id:%22RWTH-CONV-123451%22

OAI identifier oai:identifier

oai:publications.rwth-aachen.de:61830