mechanistic models offer important insights into the system [36, 37], but require
process data that is very expensive, if at all obtainable.
There is no doubt that the high complexity of living organisms, especially in
connection with their flexible information networks, which have been developed in
millions of years of evolution, is the main challenge in modeling biological processes, and hence the reason why purely in silico development methodologies are
doomed to fail [17]. Natural organisms have been so successful in surviving
unfavorable conditions as they are characterized by genetic and metabolic heterogeneity between the individuals in a population. This increases especially under
stress conditions in an effort to maximize the chance for some to survive
[38]. Bioprocesses on an industrial scale (1) contain highly heterogeneous cell
populations and (2) induce a constantly changing environment that sets the organisms under a high stress due to the inhomogeneities in the reactor. It is for this reason
that advanced experimental facilities, sensor technologies, and mathematical modeling must be tightly integrated into a digital twin framework to finally achieve
development times comparable to other industries [39–43]. The predictions at the
digital level require a continuous re-calibration and evolving mathematical description to cope with the unpredictable behavior of living systems as well as efficient
strategies to design and operate the experimental campaigns [29]. Most importantly,
these models also reflect the dynamic development of the heterogeneity of the
population and take statistical uncertainties into account in order to account for
possible batch-to-batch variations on an industrial scale and be robust in terms of
their prediction.
3 Inhomogeneities in Industrial-Scale Bioreactors
and Their Influence on the Biological System
Bioprocess development is usually started in shaken cultures, traditionally in shake
flasks, or more recently in parallel microwell plates where only a few endpoint
measurements are possible [44]. Nowadays there is a great interest in implementing
the final strategy, the fed-batch method, very early in the development process. For
this purpose, various methods have been developed in the past years, which make
this possible despite the small volumes [45].
Although this represents a significant advance on the traditional approach, the
methodologies can vary greatly depending on the applied system. Substrate feeding
can be either continuous or intermittent (pulse-based), and the controllers for the
continuously measured parameters (e.g., pH value and dissolved oxygen) can be set
differently, if these parameters are adjusted at all. If, as discussed above, we assume
that the culture is highly sensitive to the process conditions – and thus the product
formation is influenced accordingly – it is necessary to set these parameters so that
the conditions are similar to those on an industrial scale.
Potential of Integrating Model-Based Design of Experiments Approaches and. . .
7
process data that is very expensive, if at all obtainable.
There is no doubt that the high complexity of living organisms, especially in
connection with their flexible information networks, which have been developed in
millions of years of evolution, is the main challenge in modeling biological processes, and hence the reason why purely in silico development methodologies are
doomed to fail [17]. Natural organisms have been so successful in surviving
unfavorable conditions as they are characterized by genetic and metabolic heterogeneity between the individuals in a population. This increases especially under
stress conditions in an effort to maximize the chance for some to survive
[38]. Bioprocesses on an industrial scale (1) contain highly heterogeneous cell
populations and (2) induce a constantly changing environment that sets the organisms under a high stress due to the inhomogeneities in the reactor. It is for this reason
that advanced experimental facilities, sensor technologies, and mathematical modeling must be tightly integrated into a digital twin framework to finally achieve
development times comparable to other industries [39–43]. The predictions at the
digital level require a continuous re-calibration and evolving mathematical description to cope with the unpredictable behavior of living systems as well as efficient
strategies to design and operate the experimental campaigns [29]. Most importantly,
these models also reflect the dynamic development of the heterogeneity of the
population and take statistical uncertainties into account in order to account for
possible batch-to-batch variations on an industrial scale and be robust in terms of
their prediction.
3 Inhomogeneities in Industrial-Scale Bioreactors
and Their Influence on the Biological System
Bioprocess development is usually started in shaken cultures, traditionally in shake
flasks, or more recently in parallel microwell plates where only a few endpoint
measurements are possible [44]. Nowadays there is a great interest in implementing
the final strategy, the fed-batch method, very early in the development process. For
this purpose, various methods have been developed in the past years, which make
this possible despite the small volumes [45].
Although this represents a significant advance on the traditional approach, the
methodologies can vary greatly depending on the applied system. Substrate feeding
can be either continuous or intermittent (pulse-based), and the controllers for the
continuously measured parameters (e.g., pH value and dissolved oxygen) can be set
differently, if these parameters are adjusted at all. If, as discussed above, we assume
that the culture is highly sensitive to the process conditions – and thus the product
formation is influenced accordingly – it is necessary to set these parameters so that
the conditions are similar to those on an industrial scale.
Potential of Integrating Model-Based Design of Experiments Approaches and. . .
7