APR September/October 2020 - 46

MANUFACTURING

if any abnormal process behavior occurs (and if so, to what extent),
and/or if the process is under state of control, etc. It demonstrated that
modelling can not only enable the continuous prediction of otherwise
discrete off-line generated data, but also enable critical process states
being diagnosed and detected more easily. Figure 1 shows that the
prediction profiles agree well with the experimental measured profiles
for the key metabolites.

Implementation of Real-Time Release Testing
(RTRT) as part of manufacturing release strategy
As discussed previously, the RTRT concept has been applied in
regulatory approval of certain batch processes and continuous
manufacturing processes. There are several discussions on RTRT
available, including regulatory perspective69-70 and scientific
case studies from small-molecule and biopharmaceuticals
which demonstrated how RTRT can be facilitated by integrating
manufacturing and multivariable control strategies to enhance and
assure the quality of products.71-72 In principle, data streams from
an integrated manufacturing process can be continuously used
to make predictions and inform control systems. Juan L. Torres72
depicted a case study on how PAT and real-time multivariate analysis
for trending, actions and disposition decisions can be integrated to
establish drug substance advanced process controls.

Integrating PAT with data systems for
pharmaceutical manufacturing
PAT is a highly integrated cross-disciplinary area. PAT should not go
without data analysis and control algorithms, as both have been
playing unique yet critical roles in process data analysis and control
execution, including but not limited to: (1) basic multivariate analysis
chemometrics (PCA, PLS, PCR, etc.) packages have been applied to
establish multivariate models which linked raw materials' properties,
critical process parameters (CPPs), and critical quality attributes
(CQAs)73-74; (2) advanced control algorithms (e.g., feedback forward/
feedback control, automated control strategy, proportional integral
(PI) controller, etc.)75-79 have been applied for process automation,
manufacturing robustness, and continuous manufacturing; and (3)
more advanced algorithms such as pattern recognition and artificial
neural network (ANN),80 deep learning and artificial intelligence
(AI),81-85 etc., have been used to model complex patterns and problems
including predictive maintenance for electrochemical sensors,81
medical imaging processing,82 materials degradation,83 etc. However, it
should be noted that significant regulatory challenges76 for AI systems
exist, if we consider that the intent and causation may fail to function
because of the nature of the machine-learning algorithms on which
modern AI are commonly built, and that it may not be possible to
truly understand how a trained AI program is arriving at its decisions

Figure 1. The prediction results from individual PLS calibration models for each component in a bioreactor mAb IgG3 cell culture batch in
comparison with the experimental data. (a) Glucose; (d) Glutamine; (c) Lactate; (d) Ammonium.
Figure 6 from "Frontiers of Chemical Science and Engineering. Real time monitoring of bioreactor mAb IgG3 cell culture process dynamics via Fourier
transform infrared spectroscopy: Implications for enabling cell culture process analytical technology. 9(3), 2015, 386-406. H. Wu, E. Read, M. White, B. Chavez,
K. Brorson, C. Agarabi, M. Khan". "With permission of Springer Nature".

46 |

APR_SeptOct2020.indd 46

| September/October 2020

10/2/20 11:45 AM

APR September/October 2020

Table of Contents for the Digital Edition of APR September/October 2020