Advanced modeling techniques in hydrophobic interaction chromatography (HIC)

Hydrophobic Interaction Chromatography (HIC) is widely used in biopharmaceutical downstream processing for protein separation under mild and aqueous conditions. Accurate modeling of HIC for optimizing monoclonal antibody (mAb) monomer and aggregate separation remains challenging, primarily due to complex resin-protein interactions across diverse operating ranges. This study compares five advanced modeling approaches, mechanistic models, residual-based and differential parallel hybrid models, a serial hybrid model, and a physics informed neural network (PINN), to predict HIC behavior under varied conditions. Model performance was evaluated using qualitative trend analysis and quantitative metrics (R², RMSE, MBE). Among the tested approaches, the differential parallel hybrid model consistently demonstrated superior performance, accurately capturing the complex dynamics of HIC, including breakthrough, plateau, wash, and tailing phases for both monomer and aggregate species. It achieved remarkable accuracy for aggregate predictions (R² = 0.99, RMSE = 0.02, MBE = 0.01) and similarly strong performance for monomer (R² = 0.99). These findings underscore that sophisticated hybrid modeling strategies, especially the differential parallel approach, offer robust and highly accurate predictive capabilities essential for optimizing complex downstream bioprocesses like HIC.