Prediction of the solubility of medium-sized pharmaceutical compounds using a temperature-dependent NRTL-SAC model

In this work, the NRTL-SAC and the Pharma UNIFAC models are evaluated with respect to the capability of prediction of solid liquid equilibria of pharmaceutical compounds in organic solvents. The original NRTL-SAC model is extended through the introduction of temperature-dependent binary interaction parameters, and the two versions of the model are parametrized using vapor liquid equilibrium (VLE) data. The performance of the NRTL-SAC models for correlation and prediction of the solubility of eight medium-sized flexible pharmaceutical or pharmaceutically similar molecules in multiple pure, organic solvents is examined: risperidone, fenofibrate, fenoxycarb, tolbutamide, meglumine, butyl paraben, butamben, and salicylamide. The performance of the Pharma UNIFAC model is evaluated using data for six of these compounds. In general, it is found that introducing a dependence on temperature to the binary interaction parameters of the NRTL-SAC model can improve its capability for modeling and prediction of the solubility of active pharmaceutical ingredients. For prediction of solubility data the Pharma UNIFAC model generally performs below the two NRTL-SAC models. Averaged over all evaluated systems where the solubility was predicted with each method, values of the root mean squared logarithmic error (RMSLE) in predicted mole fraction solubility obtained for Pharma UNIFAC (30 systems) and for the original and the modified temperature-dependent forms of the NRTL-SAC model (29 systems) are 1.64, 1.17, and 1.09, respectively. Comparing only those systems for which all models were evaluated (18 systems), the RMSLE values are 1.42, 1.06, and 0.87, respectively.