A new chemical structure-based model to estimate solid compound solubility in supercritical CO2

Abstract

Utilization of new approaches in the determination of drug solubility in supercritical fluids can reduce the computation time and represent reliable results. This also leads to more applications of the supercritical technology in the field of drug manufacturing. A least-square support vector machine (LSSVM) approach is employed in this study in order to predict 33 different drug solubility in supercritical CO2. The solubility of the drugs is estimated as a function of temperature, pressure, supercritical CO2 density, and 20 different chemical substructures. LSSVM results are then compared to those obtained from 8 previously reported semi-empirical correlations. Satisfying predictions are performed by the proposed LSSVM with an average absolute relative deviation of 4.92% and determination coefficient of 0.998 for the testing dataset. Therefore, the proposed LSSVM can be applied as a reliable predictive tool to estimate the drugs’ solubility, if drugs’ chemical structures are given.