CPV of the Future: AI-Powered Continued Process Verification for Bioreactor Processes.

Area of Science:

• Biopharmaceutical Manufacturing
• Process Validation
• Machine Learning Applications

Background:

• Biopharma process validation follows FDA guidelines: Process Design (PD), Process Qualification (PQ), and Continued Process Verification (CPV).
• CPV analytics require extensive process knowledge, often unavailable for new drugs and processes.
• Limited historical data poses challenges for validating analytical methods in new bioprocesses.

Purpose of the Study:

• To evaluate the suitability of machine learning/artificial intelligence (ML/AI) methods for Continued Process Verification (CPV) in bioprocesses.
• To assess ML/AI models for real-time monitoring and cell physiological control of yeast Pichia pastoris.
• To demonstrate the application of ML/AI in a case study of recombinant lipase 1 (Crl1) production under hypoxic conditions.

Main Methods:

• Utilized supervised and unsupervised machine learning models with historical data from fed-batch bioprocesses.
• Applied a multivariate anomaly detection (isolation forest) model to the batch phase.
• Assessed a supervised random forest model for predicting operator control actions during the fed-batch phase to maintain respiratory quotient (RQ).

Main Results:

• The isolation forest model effectively detected anomalies in the bioprocess batch phase.
• The random forest model accurately predicted operator control actions for maintaining RQ within the desired range.
• Models were validated using subject matter expert evaluation and real-time data implementation.

Conclusions:

• Machine learning-based multivariate analytics are suitable for Continued Process Verification (CPV) in biopharma manufacturing.
• ML/AI tools can enhance real-time monitoring and control of bioprocesses, improving efficiency and product quality.
• This study provides a proof-of-concept for using ML/AI with limited data in biopharmaceutical process validation.