Accelerating mechanistic model calibration in protein chromatography using artificial neural networks

Area of Science:

• Biopharmaceutical Manufacturing
• Process Development
• Chromatography

Background:

• Mechanistic models are crucial for optimizing chromatography in therapeutic monoclonal antibody (mAb) manufacturing.
• Model calibration is a significant bottleneck, hindering industrial implementation.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN)-assisted workflow for accelerated mechanistic model calibration.
• To improve the efficiency of parameter estimation for the steric mass-action (SMA) isotherm model in cation exchange chromatography (CEX).

Main Methods:

• A semi-automated ANN-assisted calibration workflow was developed.
• The workflow was applied to calibrate the SMA isotherm model for CEX using two mAb feedstocks of varying complexity.
• Different training data combinations and parameter estimation strategies (one-step vs. two-step) were investigated.

Main Results:

• The ANN-assisted workflow yielded acceptable parameter estimations and good agreement between experimental and simulated chromatograms for target compounds.
• A two-step parameter estimation approach using high-load data improved impurity prediction for complex feedstocks.
• The workflow significantly reduced calibration effort, especially for complex mAb processes.

Conclusions:

• ANN-assisted calibration offers a promising approach to streamline mechanistic model implementation in biopharmaceutical process development.
• This method enhances the efficiency of downstream process development for therapeutic antibodies, particularly those with complex structures and impurity profiles.