Computational prediction of drug solubility in supercritical carbon dioxide: Thermodynamic and artificial intelligence modeling

Abstract

In this study, machine learning (ML) computations were carried out for description of drug solubility in supercritical carbon dioxide. Supercritical solvent has been used in this work due to its superior properties and high solvation capacity for drug dissolution. The model has been developed and tested for salsalate as well as decitabine drugs, and their solubility at various pressures and temperatures were predicted using the developed machine learning model. The models were developed by taking into account the pressure between 120 and 400 bar, and temperature between 308 and 338 K for understanding the influence of pressure and temperature on salsalate and decitabine drug dissolution in the solvent. Moreover, the model’s accuracy was compared with some empirical correlations from previous studies. It was indicated that the ML model had better accuracy compared to the semi-empirical correlations. The pressure was indicated to have considerable influence on the solubility variations for both drugs. The best thermodynamic model showed the least average absolute relative deviation percent of around 8 % for the whole data points for salsalate. For development of machine learning model, artificial neural network was trained using the measured data. The neural network was developed using one hidden layer, two inputs, and one output. Pressure and temperature were taken as inputs for the network, and the solubility of drug as the predicted output in the neural network. The training and validation of the neural network using salsalate and decitabine solubility indicated great accuracy with coefficient of determination higher than 0.99 for both steps.