An Advanced Monitoring Platform for Rational Design of Recombinant Processes - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 76

3 Chemometry and Prediction
The process monitoring platform described above delivers comprehensive online
and offline datasets, but although advanced sensor systems are employed, direct
measurement of key biological process variables (cell dry mass, product titer, and
stress markers) is not possible. The problem is that physiology-relevant informa-
tion delivered by offline analytics is not accessible in real time and most of the
online sensors reflect changes in the process not directly linked to biological
process variables per se. Statistical and mathematical modeling is indispensable to
bridge this gap to key physiology-relevant variables to improve understanding of
the process [ 57 ]. The application of mathematical methods, including multivariate
data analysis, chemometrics, and statistical methods, enables the extraction of
meaningful information from a variety of signals from online sensors and their
assignment to complex offline variables.
The applicability of this approach has been demonstrated for the prediction of
biomass, TCN, recombinant protein content, and plasmid copy number in E. coli
cultivation processes. PLS regression and a radial basis function neural network
(RBF-NN) model were tested for their predictive power. The best results were
obtained with the RBF network model with a selected set of online data highly
correlated to offline variables (Fig. 4 ).
Fig. 4 Prediction of target process variables (CDM, TCN, DC, product, and PCN) by a RBF
model generated from selected input signals