Solubility prediction of pharmaceuticals in dioxane + water mixtures at various temperatures: Effects of different descriptors and feature selection methods

Abstract

Solubility of drugs in dioxane + water mixtures at different temperatures was mathematically described by the Jouyban-Acree model employing different solute descriptors; i.e. 1) Abraham solvation, 2) partial solubility, 3) HyperChem and 4) Dragon descriptors in the solute-solvent interactions terms to represent the effects of solute structure on the solubility of drugs in dioxane + water as a model solvent mixture system. Trained versions of these models using collected solubility data in dioxane + water mixtures were proposed and their accuracies were evaluated employing the individual percentage deviation (IPD) and the mean of IPD (MPD) for each drug. Different feature selection methods were applied to select the relevant descriptors and the accuracy of the different descriptors and these approaches were compared. The validity of each model was evaluated using the leave-one-drug out validation method. The results of Dragon descriptors selected by the enhanced replacement method gave a significantly improved model for predicting solubility in dioxane + water. The results confirm that solute-solvent interaction is a critical factor in predicting the solubility of drugs in solvent mixtures. In addition, the feature selection is a critical issue in developing a quantitative structure-property relationships (QSPR) model. © 2014 Elsevier Ltd. All rights reserved.