Advances in industrial biopharmaceutical batch process monitoring: Machine-learning methods for small data problems

Area of Science:

• Biopharmaceutical Manufacturing
• Process Analytical Technology (PAT)
• Machine Learning in Quality Control

Background:

• Real-time multivariate statistical batch process monitoring (BPM) is crucial for ensuring quality and detecting deviations in biopharmaceutical manufacturing.
• The "Low-N" problem, characterized by limited production history, hinders the effective deployment of multivariate BPM.
• Current industry practice involves switching to univariate BPM, risking the detection of critical process deviations.

Purpose of the Study:

• To address the industry-wide "Low-N" problem in biopharmaceutical batch process monitoring.
• To propose and validate a novel approach for generating synthetic batch data to overcome limitations of small historical datasets.
• To enable robust multivariate BPM even with limited product history.

Main Methods:

• Development of a machine learning-based methodology combined with hardware exploitation to generate a large number of in silico (simulated) batches.
• Application of the proposed approach to address the "Low-N" challenge in batch process monitoring.
• Validation through several industrial case studies in bulk drug substance manufacturing.

Main Results:

• Successful generation of an arbitrarily large number of in silico batches, effectively simulating extensive production history.
• Demonstration of the proposed approach's efficacy in diverse Low-N scenarios within bulk drug substance manufacturing.
• Overcoming the limitations of traditional univariate monitoring for processes with insufficient historical data.

Conclusions:

• The proposed machine learning and hardware exploitation method offers a viable solution to the "Low-N" problem in biopharmaceutical BPM.
• This innovative approach enhances process monitoring capabilities, mitigating risks associated with limited historical data.
• This represents the first application of machine learning to solve the "Low-N" problem in this manufacturing sector.