Hybrid Dynamic Modelling With Gaussian Process Regression for Intensified Fed-Batch CHO Cell Culture Processes

Area of Science:

• Biotechnology
• Process Engineering
• Computational Biology

Background:

• Increasing complexity in cell culture demands advanced process modeling for efficient biomanufacture.
• Hybrid models combining machine learning and material balances offer a path to digital bioprocess development.

Purpose of the Study:

• To propose and evaluate a hybrid dynamic model using Gaussian Process Regression (GPR) for predicting cell growth and metabolism.
• To compare the GPR-hybrid model's predictive performance against a Monod-based kinetic model.
• To investigate the impact of data availability and quality on GPR-hybrid model performance across development and manufacturing scales.

Main Methods:

• Development of a hybrid dynamic model integrating Gaussian Process Regression (GPR) with material balance principles.
• Evaluation of the GPR-hybrid model on two intensified fed-batch Chinese Hamster Ovary (CHO) cell culture processes.
• Comparative analysis of predictive performance against a standard Monod-based kinetic model.

Main Results:

• The GPR-hybrid model demonstrated superior predictive capabilities compared to the Monod-based approach.
• Data quality significantly impacts the predictive accuracy of GPR-hybrid models.
• Model performance was assessed across both development and manufacturing scales, showing scalability.

Conclusions:

• Hybrid GPR models show significant promise for data-driven bioprocess development, outperforming mechanistic models.
• Emphasizes the critical role of data quality and experimental design in ensuring model reliability.
• Identified challenges with transient feeding effects warrant further investigation for enhanced model structure.