Model-based process development for hydrophobic interaction chromatography by considering prediction uncertainty...

Area of Science:

• Biotechnology
• Chemical Engineering
• Computational Modeling

Background:

• Mechanistic models are crucial for hydrophobic interaction chromatography (HIC) process development.
• Parameter estimation can calibrate models, but inherent biases limit prediction accuracy.
• Discrepancies between predicted (97.3%) and experimental (86.0%) yields highlight model limitations.

Purpose of the Study:

• To address model prediction biases in HIC process optimization.
• To quantify uncertainty in model parameters and predictions.
• To develop a framework for risk-averse, uncertainty-informed process development.

Main Methods:

• Employed Bayesian inference with Markov Chain Monte Carlo (MCMC) for parameter uncertainty calculation.
• Transformed parameter uncertainty into model prediction uncertainty.
• Integrated uncertainty analysis into HIC process optimization.

Main Results:

• Identified significant yield discrepancies in a well-calibrated HIC model.
• Quantified model-predicted yield uncertainty (76.9%–96.5%), aligning with experimental observations.
• Re-optimized process achieved narrower yield uncertainty (94.2%–98.9%) and high experimental yield (95.8%).

Conclusions:

• Uncertainty quantification in HIC models improves the reliability of process optimization.
• This approach aids in reflecting model prediction deviations and reducing development risks.
• A proposed framework enhances model prediction accuracy and minimizes risks in model-based process development.