Experimental enhancing drug solubility in supercritical CO₂ using modifiers: Modeling and machine learning prediction

The experimental determination of the solubility of aripiprazole (APZ) in supercritical carbon dioxide (SC-CO 2 ) containing ethanol and dimethyl sulfoxide (DMSO) was conducted at temperatures ranging from 308 to 338 K and pressures from 12 to 30 MPa. The measured mole fraction solubility of ethanol ranged from 2.95 × 10 − 5 to 1.03 × 10 − 4 , while that of DMSO ranged from 4.59 × 10 − 5 to 1.38 × 10 − 4 . These results demonstrate that DMSO provides superior solubility enhancement. The maximum enhancement factor achieved was approximately 33 relatives to pure SC-CO₂. The experimental were then correlated using six semi-empirical models: The Sodeifian-Sajadian, González et al., Jouyban et al., Garlapati-Madras, MST, and Soltani-Mazloumi models. The Sodeifian-Sajadian and González et al. models were found to give the best correlations in terms of AARD%, R adj, and AICc, respectively, for DMSO and ethanol. In addition, multilayer perceptron (MLP) and support vector regression (SVR) models were developed to predict APZ solubility. The MLP model demonstrated superior predictive capabilities, with AARD values of 0.85% and 1.30% for the ethanol and DMSO systems, respectively. The experimental data and predictive models provide a reliable foundation for designing and optimizing supercritical pharmaceutical particle engineering processes.