Computer aided model based bioprocess development by automated on-line optimal re-design in parallel bioreactors

Mini-bioreactors and their integration into robot systems speed-up bioprocess development by enabling parallel fed-batch experiments. They also support model-based bioprocess development to further reduce experimental costs. Therefore, mathematical methods are required to compute best experimental set-up´s aiming to fit an existing model with minimal effort. Statistical methods for Design of Experiments using multivariate regression models have already been implemented in high-throughput-screening, but are not suitable for fed-batch cultivations, which are typically described by nonlinear differential equation models. Methods for Optimal Experimental Design (OED), however, can deal with the nonlinear systems.

Here a Sliding-Window Optimal Experimental Re-Design (SWORD) is applied to eight parallel Escherichia coli fed-batch cultivations, performed in a sophisticated robot station, which also performs automatically all needed analyses at-line during the experiment. This allows to re-fit the model after each sample in order to re-design the experiment using the information that has been generated throughout the running experiment.

The results show that the on-line computation of the optimal setup for all parallel experiments is possible by to the proper length selection of the sliding window together with efficient treatment of the underlying nonlinear parameter estimation and experiment design problems. SWORD reduces the confidence interval of the parameter set significantly and increases robustness and flexibility of the experimental design through its recursive re-design. As a result, the parameters for a complex model with many differential equations can be identified in a single experiment, which drastically reduces time and labor and provides a realistic basis towards a computation based bioprocess development.