mixing effects in the scale-down bioreactor. The authors reported significant yield
losses, incorporation of non-conventional amino acids into the recombinant protein
product and accumulation of metabolites in response to the calculated gradients.
Although these results were similar to observations in other non-model-based
scale-down approaches [69, 102, 103], the added advantage of using the model was
the flexibility of stress definition, in which one parallel experimental set-up was used
to implement six different stress zones in the scale-down system, which otherwise
could not be done manually.
A further advancement of this concept is the application of models to design
intelligent scale-down experiments. In all the previously discussed scale-down
methods, the researcher sets out a prior design space, within which the stresses
and gradient profiles are pre-defined (manually or using a model) and executed
during the experiment. The use of digital twins to advance scale-down design should
involve designing the stresses as the experiment runs. In other words, the nature of
gradients to be imposed on the culture at time point t2 will depend on its state at time
point t1. This will prevent overestimating or underestimating the magnitude of
gradients, as well as the exposure time of the cells to a given gradient. Such an
adaptive re-design technique was employed by Cruz Bournazou and colleges in
designing optimal experiments to maximize information content for model parameter identification [40, 104]. Although no scale-down efforts were made in these
works, the authors demonstrated the ability to re-define feed regimes based on the
current state of the culture and model predictions for a given time window. Such a
model-based adaptive system can easily be employed to execute dynamic scaledown experiments.
The addition of digital twin concepts to the definition of the scale-down model
offers flexibility of stress definition, automation, and a high turnover in experimental
throughput, to drive the digital revolution in bioprocess engineering, as discussed by
Neubauer et al. [30]. The few pioneering works published so far point in the future
direction where mathematical methods, in the form of digital twins, will help to
design more informative and smart (scale-down) experiments, to move away from
the traditional, commonly used static design of experiment (DoE) paradigm.
Model-Based Interpretation of Scale-Down Data
Scale-down bioreactors offer important insights into cellular behavior under heterogeneous fermentation conditions of larger-scale bioreactors. The data is usually
interpreted at the macroscopic level, by comparing metabolite, substrate, and growth
profiles to cultivations under homogeneous conditions. In a few studies, derivative
indices (e.g., specific uptake rates, yield coefficients) have been calculated from the
raw scale-down data to support the interpretation of the data, e.g. in [105]. A few
extensions of the data space in scale-down experiments involve molecular level
analysis. For instance, Simen and co-workers used transcriptomics data from an
STR-PFR scale-down bioreactor to monitor different gene expression levels under
short-term and long-term substrate fluctuations in E. coli culture [58]. In another
Potential of Integrating Model-Based Design of Experiments Approaches and. . .
19
losses, incorporation of non-conventional amino acids into the recombinant protein
product and accumulation of metabolites in response to the calculated gradients.
Although these results were similar to observations in other non-model-based
scale-down approaches [69, 102, 103], the added advantage of using the model was
the flexibility of stress definition, in which one parallel experimental set-up was used
to implement six different stress zones in the scale-down system, which otherwise
could not be done manually.
A further advancement of this concept is the application of models to design
intelligent scale-down experiments. In all the previously discussed scale-down
methods, the researcher sets out a prior design space, within which the stresses
and gradient profiles are pre-defined (manually or using a model) and executed
during the experiment. The use of digital twins to advance scale-down design should
involve designing the stresses as the experiment runs. In other words, the nature of
gradients to be imposed on the culture at time point t2 will depend on its state at time
point t1. This will prevent overestimating or underestimating the magnitude of
gradients, as well as the exposure time of the cells to a given gradient. Such an
adaptive re-design technique was employed by Cruz Bournazou and colleges in
designing optimal experiments to maximize information content for model parameter identification [40, 104]. Although no scale-down efforts were made in these
works, the authors demonstrated the ability to re-define feed regimes based on the
current state of the culture and model predictions for a given time window. Such a
model-based adaptive system can easily be employed to execute dynamic scaledown experiments.
The addition of digital twin concepts to the definition of the scale-down model
offers flexibility of stress definition, automation, and a high turnover in experimental
throughput, to drive the digital revolution in bioprocess engineering, as discussed by
Neubauer et al. [30]. The few pioneering works published so far point in the future
direction where mathematical methods, in the form of digital twins, will help to
design more informative and smart (scale-down) experiments, to move away from
the traditional, commonly used static design of experiment (DoE) paradigm.
Model-Based Interpretation of Scale-Down Data
Scale-down bioreactors offer important insights into cellular behavior under heterogeneous fermentation conditions of larger-scale bioreactors. The data is usually
interpreted at the macroscopic level, by comparing metabolite, substrate, and growth
profiles to cultivations under homogeneous conditions. In a few studies, derivative
indices (e.g., specific uptake rates, yield coefficients) have been calculated from the
raw scale-down data to support the interpretation of the data, e.g. in [105]. A few
extensions of the data space in scale-down experiments involve molecular level
analysis. For instance, Simen and co-workers used transcriptomics data from an
STR-PFR scale-down bioreactor to monitor different gene expression levels under
short-term and long-term substrate fluctuations in E. coli culture [58]. In another
Potential of Integrating Model-Based Design of Experiments Approaches and. . .
19