Development and evaluation of artificial neural network and hybrid PR-ANN models for accurate solubility prediction of Amoxapine in supercritical carbon dioxide

Abstract

The importance of prediction of drugs solubility in supercritical carbon dioxide (SC-CO2) green solvent is crucial in order to micro or nanosizing the drug particles in pharmaceutical industry. In this study, an artificial neural network (ANN) model as well as a hybrid model that used the Peng- Robinson equation of state with ANN (PR-ANN) was used for for accurate solubility prediction of Amoxapine in SC-CO2. The ANN only performed adequately as a predictor (AARD=8.05%, R2 = 0.976), while the hybrid PR-ANN accurately predicted Amoxapine solubility over a range of pressures and temperatures (AARD= 2.04%, R2 =0.998). To verify that the model was accurately capturing all the parameters related to solubility prediction, five-fold cross-validation has demonstrated its generalizability with mean values of AARD and R2 =8.46% and 0.956, respectively. Thus, these results indicate that the hybrid PR-ANN provides reliable predictions of the solubility of Amoxapine that includes both pressure and temperature dependency.