Advanced modeling techniques in hydrophobic interaction chromatography (HIC).

Area of Science:

• Biopharmaceutical downstream processing
• Chromatographic separation science
• Protein aggregate analysis

Background:

• Hydrophobic Interaction Chromatography (HIC) is crucial for biopharmaceutical protein purification.
• Accurate modeling of HIC is challenging due to complex resin-protein interactions.
• Optimizing monoclonal antibody (mAb) monomer and aggregate separation requires robust predictive tools.

Purpose of the Study:

• To compare five advanced modeling approaches for predicting HIC behavior.
• To evaluate model performance using qualitative and quantitative metrics.
• To identify the most effective modeling strategy for HIC optimization.

Main Methods:

• Comparison of mechanistic models, residual-based hybrid models, differential parallel hybrid models, serial hybrid models, and physics-informed neural networks (PINNs).
• Evaluation of model accuracy using R², Root Mean Square Error (RMSE), and Mean Bias Error (MBE).
• Analysis of HIC phases including breakthrough, plateau, wash, and tailing for monomer and aggregate species.

Main Results:

• The differential parallel hybrid model demonstrated superior performance across varied conditions.
• This model accurately predicted HIC dynamics for both mAb monomers and aggregates.
• Exceptional accuracy was achieved for aggregate prediction (R² = 0.99, RMSE = 0.02, MBE = 0.01) and monomer prediction (R² = 0.99).

Conclusions:

• Sophisticated hybrid modeling, particularly the differential parallel approach, offers highly accurate predictive capabilities for HIC.
• This modeling strategy is essential for optimizing complex downstream bioprocesses.
• The findings support the use of advanced hybrid models for enhanced bioseparation efficiency and product quality.