Prediction uncertainty assessment of chromatography models using Bayesian inference

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Pharmaceutical Process Development

Background:

• Mechanistic modeling in chromatography is established in academia but underutilized in pharmaceutical industry.
• Hesitation stems from a lack of standardized methods for model qualification in process development.

Purpose of the Study:

• Introduce a Bayesian framework for calibrating and assessing mechanistic chromatography models.
• Provide a general approach for determining model suitability for decision-making in pharmaceutical process development.

Main Methods:

• Utilize Bayesian Markov Chain Monte Carlo (MCMC) to estimate parameter posterior distributions and assess uncertainty.
• Propagate parameter uncertainty to model predictions for a robust uncertainty assessment.
• Apply the framework to a model describing antibody separation from impurities on a strong cation exchanger.

Main Results:

• Demonstrate the framework's ability to quantify parameter uncertainty.
• Showcase successful extrapolation of a model calibrated at moderate load density to high load conditions.
• Validate the model's qualification for supporting process development despite parameter uncertainty.

Conclusions:

• The proposed Bayesian framework effectively qualifies mechanistic chromatography models for pharmaceutical applications.
• This approach enables reliable model-based predictions and decision-making, even for extrapolations.
• Systematic application of this framework can enhance pharmaceutical process development efficiency and reliability.