Multiblock PLS analysis of an industrial pharmaceutical process

Area of Science:

• Pharmaceutical Process Engineering
• Chemometrics
• Biotechnology

Background:

• Industrial production of active pharmaceutical ingredients (APIs) via fermentation requires robust process monitoring and control.
• Traditional multivariate data analysis methods may not fully capture the complexity of multi-stage bioprocesses.

Purpose of the Study:

• To model an industrial API fermentation process using multiblock partial least squares (MBPLS).
• To identify key process stages and variables influencing API production.
• To compare MBPLS with standard partial least squares (PLS) for process analysis.

Main Methods:

• Application of multiblock partial least squares (MBPLS) to model fermentation data from 30 batches.
• Segregation of data into four blocks: inoculum quality, inoculum manipulated, API quality, and API manipulated variables.
• Development of a reduced variable set model based on initial analysis.

Main Results:

• MBPLS successfully modeled the API fermentation process, differentiating contributions from inoculum and API production stages.
• Inoculum quality variables strongly correlated with API production in nominal fermentations.
• Fermentation manipulation variables (pH, biomass) were critical for API yield in non-nominal runs.
• A reduced model achieved 82.4% variance prediction for final API concentration via cross-validation.

Conclusions:

• MBPLS offers advantages over standard PLS by dissecting process stage contributions to API volumetric productivity.
• The study identified critical control points for optimizing API fermentation performance.
• The developed model provides a basis for improved process understanding and control.