Prediction of centrifugal partition chromatography separation performance via statistical design

Centrifugal Partition Chromatography (CPC) is widely used for purifying natural products and fine chemicals due to its high selectivity and full product recovery. However, transferring separations from analytical to preparative scale often leads to deviations from ideal liquid-liquid partitioning, where retention and band broadening can no longer be described solely by partition coefficients and hydrodynamic residence time. Additional effects related to operating conditions and hydrodynamics become relevant. In this work, a data-driven modeling approach is applied to quantify how operating parameters and solute properties affect CPC performance in the biphasic system Arizona N. A full-factorial Design of Experiments was conducted across three rotor volumes (99 mL, 154 mL, 204.3 mL) in ascending and descending modes. The three investigated factors included mobile phase flow rate, sample volume and global concentration for varied partition coefficients, while retention time and peak width served as response variables. Box-Cox transformations with maximum likelihood estimation were used to improve model adequacy prior to regression. The resulting models include significant linear, interaction, and quadratic terms, enabling empirical description of deviations from ideal partitioning behavior. The models show strong predictive performance (R² and Q² > 0.95), with most predictions within ± 20 % of experimental values. Retention is mainly governed by flow rate, rotor volume, and partition coefficient, while peak width is additionally influenced by injection volume and interactions. Quadratic contributions indicate non-linear dependencies, potentially linked to hydrodynamic or mass-transfer effects, although their mechanistic origin cannot be conclusively resolved within the scope of this study. Overall, this work provides a first quantitative framework for describing CPC performance across different rotor scales under preparative conditions and highlights the potential of empirical, data-driven approaches as an initial step toward systematic understanding, optimized operation, and future closed-loop process control in preparative CPC.