Applying Mechanistic Models in Bioprocess Development - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 164

replicates (duplicate/triplicate fermentations) but also sample replicates are needed
to know the error of the measurements. If the quality of the collected data is not
sufficiently high, this might later raise severe questions about the reliability of the
resulting model.
Assuming that a decision has been taken to develop a mechanistic model of a
pharmaceutical production process, or one of its unit operations, one could, of
course, wonder how such a model can be established, and how it can support PAT
objectives. In general, construction of a mechanistic model is considered time-
consuming, which may explain why data-driven models and chemometrics have
been more popular than mechanistic approaches, despite the PAT guidance. How-
ever, during the past 5 years, this situation has already changed considerably for
small-molecule drug substances [ 4 ]. According to us, the tools presented here can be
helpful in setting up and structuring the model equations in an efficient way, for
example, by making use of matrix notation, which can facilitate transfer of the model
equations between different users. Such sharing of modeling knowledge is essential
in multidisciplinary process development. As discussed by Sin et al [ 14 ], a signif-
icant part of such a model matrix can be transferred from one system to a second or a
third, which undoubtedly makes the whole model-building exercise more efficient.
Finally, we would also like to emphasize that one should move ahead in small
steps when constructing a mechanistic model of a process or unit operation. One
should rather start with a smaller model with limited scope, for example, an
unstructured model [ 21 ]. Such a model could then be gradually extended with more
detail, while the development of the production process at laboratory and pilot scale
is ongoing. The model analysis tools presented here can then be used in the different
stages of the model-building as continuous quality checks of the model.
Once a model is considered ready for use, a first application that is relevant for
such a model is to use simulations to propose more informative experiments
leading to more accurate estimation of the model parameters, for example, by
applying optimal experimental design (OED) [ 22 ]. Furthermore, the mechanistic
model can be helpful in process design, optimization, and in development of
suitable control strategies [ 23 ]. The latter applications of the model are essential
for implementing PAT principles, and can potentially contribute to more efficient
process development, replacing data collection and experiments by simulations
whenever possible.
4 Conclusions
Mechanistic models form an attractive alternative for structuring and representing
process knowledge, also for production processes in biotechnology. The reliability
of such models can be confirmed by performing identifiability, uncertainty, and
sensitivity analyses on the resulting model. Tools for performing such analyses can
be considered as standard engineering tools and are increasingly available on