RESUMEN
The Abraham and NRTL-SAC semipredictive models were employed to represent the solubility of (-)-borneol, (1R)-(+)-camphor, l-(-)-menthol, and thymol in water and organic solvents, using data measured in this work and collected from the literature. A reduced set of solubility data was used to estimate the model parameters of the solutes, and global average relative deviations (ARDs) of 27% for the Abraham model and 15% for the NRTL-SAC model were obtained. The predictive capability of these models was tested by estimating the solubilities in solvents not included in the correlation step. Global ARDs of 8% (Abraham model) and 14% (NRTL-SAC model) were obtained. Finally, the predictive COSMO-RS model was used to describe the solubility data in organic solvents, with ARD of 16%. These results show the overall better performance of NRTL-SAC in a hybrid correlation/prediction approach, while COSMO-RS can produce very satisfactory predictions even in the absence of any experimental data.
RESUMEN
Countercurrent and centrifugal partition chromatography are techniques applied in the separation and isolation of compounds from natural extracts. One of the key design parameters of these processes is the selection of the biphasic solvent system that provides for the adequate partitioning of the solutes. To address this challenging task, the fully predictive Conductor-like Screening Model for Real Solvents (COSMO-RS) and the semi-predictive Non-Random Two-Liquid Segment Activity Coefficient (NRTL-SAC) model were applied to estimate the partition coefficients (K) of four model phenolic compounds (vanillin, ferulic acid, (S)-hesperetin and quercetin) in different solvent systems. Complementing the experimental data collected in the literature, partition coefficients of each solute in binary, or quaternary, solvent systems were measured at 298.2 K. Higher deviations from the experimental data were obtained using the predictive COSMO-RS model, with an average RMSD (root-mean-square deviation) in log(K) of 1.17 of all four solutes (61 data points), providing a satisfactory quantitative description only for the systems containing vanillin (RSMD = 0.57). For the NRTL-SAC model, the molecular parameters of the solutes were initially calculated by correlating a set of K and solubility (x, in mole fraction) data (16 partition coefficients and 44 solubility data points), for which average RMSD values of 0.07 and 0.41 were obtained in log(K) and log(x), respectively. The predictions of the remaining log(K) data (45 partition coefficients) resulted in an average RMSD of 0.43, suggesting that the NRTL-SAC model was a more reliable quantitative solvent screening tool. Depending on the amount of available solubility and partition data, both models can be valuable alternatives in the preliminary stages of solvent screening destined to select the optimal mobile and stationary phases for a given separation.