“Iterative Design of Dynamic Experiments in Modeling for Optimization of Innovative Bioprocesses”

MARIANO D. CRISTALDI; RICARDO J. A. GRAU; ERNESTO C. MARTINEZ

Process Modeling & Control-A Special Issue of Chemical Product and Process Modeling

The Berkeley Electronic Press

Año: 2009 vol. 4 p. 6 - 34

1934-2659

Today, many pharmaceuticals are produced using genetically modified microorganisms. Since the firstgene cloning in the beginning of the 1970s, it is possible to modify a microorganism to produce a desired substance, often a protein. Recombinant DNA techniques permit the creation of exactly known changes in the DNA structure. The technique is also used to manipulate the protein structures to give them better properties which has opened new possibilities to improve production strains and to produce new high valued-added biotech products. The recombinant organism is then grown in a bioreactor to large numbers to obtain considerable quantities of the target protein. The environment inside the bioreactor should allow optimal growth and product synthesis. Considering the large uncertainty and variability present in novel bioprocesses, the development of an accurate mathematical representation of bioreactor operation is a costly and difficult undertaking. One central concern is how process modeling can be best pursued considering poor knowledge about phenomena involved, sparse and biased measurements of key chemical variables and uncontrollable variations in bioreactor behavior from batch to batch. As a result, a model for a bioprocess cannot be entirely knowledgedriven or data-driven alone. Modeling for optimization is a systematic way of combining scarce data with simple first-principles models to reduce extrapolation errors in experimental optimization. In order to achieve the goal of optimal operation of innovative processes in the face of uncertainty, a number of requirements are imposed on modeling for optimization to make an impact on industrial practice. A crucial issue is how to design a rather short sequence of dynamic experiments that are most informative in order to reduce the uncertainty about the parameters of the optimal operating policy. Despite the importance of this problem there is no previous work on the development of experimental design techniques addressing the more specific objective of modeling for optimization of bioreactors during the scaling up of innovative products. The problem addressed in this work is formulated as: How does one adjust the time-varying controls, initial conditions and length of each dynamic experiment to generate the maximum amount of information for the purpose of significantly reducing the uncertainty regarding the location of the process optimum operating condition in a bioreactor? The notion of dynamic experiments highlighted the fact that some control variables are time-varying during the xperiment which rules out using standard experimental design techniques.