Optimal design of dynamic experiments in the development of cybernetic models for bioreactors

Abstract

Cybernetic models of bioreactors are appealing due to their capacity to account for regulatory mechanisms in cell metabolism by modeling the synthesis of enzymes and their activities. For a given objective of interest, experimental data used to fit the cybernetic model parameters should be maximally informative. To excite purposefully the most relevant metabolic pathways, a dynamic experiment is designed by accounting for the sensitivity of the chosen objective to time-varying operating conditions. In this work, the bioreactor feeding profile and sampling times are designed to maximize the information content. A Bayesian optimization approach is proposed to solve the resulting mathematical program. As a case study, biomass production is used as the objective to be maximized in fed-batch cultivation of Saccharomyces cerevisiae growing on glucose as a carbon source. Experimental results demonstrate that the proposed approach helps to iteratively improve a cybernetic model by designing experiments that maximize the information content.