Applying Mechanistic Models in Bioprocess Development - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 140

2 Case Study: Aerobic Cultivation of Budding Yeast

Saccharomyces cerevisiae is one of the most relevant and intensively studied
microorganisms in biotechnology and bioprocess engineering; For example, out of
151 recombinant biopharmaceuticals that had been approved by the FDA and
EMEA in January 2009, 28 (or 18.5 %) were produced in S. cerevisiae [ 13 ].
Sonnleitner and Käppeli [ 1 ] proposed a widely accepted mechanistic model
describing the aerobic growth of budding yeast, and this model is used here to
exemplify how a mechanistic model of a bioprocess can be applied to create more
in-depth process knowledge. Optimally, the process knowledge should be trans-
lated into a mechanistic model, and the model should be updated whenever
additional details of the process are unraveled. This model should capture the key
phenomena taking place in the process, and be further employed in the develop-
ment of process control strategies.
However, when developing and using mechanistic models, reliability of the
model (hence the credibility of model-based applications) is an important issue,
which needs to be assessed using appropriate methods and tools including iden-
tifiability, sensitivity, and uncertainty analysis techniques. Unfortunately, literature
reporting on mechanistic model developments often lacks the results of such
analysis—confidence intervals on estimated parameters, for example, are only
sporadically reported—and as a consequence it is not possible to conclude about
the quality of the model and its predictions. Seen from a PAT perspective, it is of
utmost importance to document that one has constructed a reliable mechanistic
model; For example, in case this model would be used later for simulations to help
in determining where to put the borders of the design space, it would be difficult to
defend the resulting design space—for example, towards the FDA—in case the
reliability of the model cannot be documented sufficiently.
One of the challenges in modeling is the identifiability problem, defined as
''given a set of data, how well can the unknown model parameters be estimated,
hence identified.'' Typically, the number of parameters in a mechanistic model is
relatively high, and therefore it is often not possible to uniquely estimate all the
parameters by fitting the model predictions to experimental measurements. An
indication of the parameters that can be estimated based on available data can be
obtained by performing an identifiability analysis prior to the parameter

estimation.
Furthermore, the model predictions will depend on the values of all parameters.
Some of the parameters will, however, have a stronger influence than others. An
uncertainty and sensitivity analysis can be performed to determine which are the
parameters whose variability contributes most to the variance of the different
model outputs.
In this case study, a systematic model analysis is performed following the
workflow presented in Fig. 1 . This workflow is rather generic, and could easily be
transferred to another case study with a similar model.