Multivariate Data Analysis for Advancing the Interpretation of Bioprocess Measurement and Monitoring Data - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 184

Fig. 10 Comparison of optimised (red) and standard (blue) settings of operating conditions for
three chromatography columns in the recovery of an industrially relevant recombinant protein
4.2.2 Process Improvement Case Study
The importance of overall process understanding and optimisation is growing.
In the case study reported here, manufacturing process data from a series of
downstream processing units of an industrially relevant recombinant protein
process were used to demonstrate that real-time optimisation of operating condi-
tions in response to the process behaviour in upstream units can lead to significant
improvements in overall product yield and recovery.
A series of three chromatographic columns with typical process data collected
from such processes were used to develop PLS and O-PLS models predicting
either product recovery or quality. The predictions of the best performing models
were subsequently used in an optimisation scheme where the operational settings
of the subsequent columns were adjusted within a pre-specified operating window
to maximise the recovery of the product whilst maintaining the required levels of
product quality.
Following offline development of the modelling and optimisation scheme using
historical process data, pilot-scale experiments with online monitoring and opti-
misation were performed. Figure 10 illustrates the comparison of two validation
batches that used optimised settings for the manipulated variables for each of the
columns with two batches running with standard operating condition settings for
each of the columns.
Clearly, significant improvements can be observed in the recovery of the
product in each of the batches used for comparison. The benefit of using a well-
established modelling methodology, on which this optimisation scheme is based, is
that in the highly regulated biopharmaceutical industry such an approach provides
a more straightforward route to implementation within a validated environment.
A similar approach of overall process optimisation based on soft sensors
combined with mechanistic models of individual unit operations has been reported
by Gao et al. [ 13 ]. The multi-agent system reported by the authors comprised a