Abstract
•Development of machine learning models for prediction of Busulfan solubility.•Estimating solubility of drug in supercritical solvent.•Evaluating effect of pressure and temperature on solubility using the models.
In this research, we used two support vector-based models to predict the performance of a supercritical-based technology in nanomedicine preparation for advanced pharmaceutical manufacturing. The models chosen are linear SVM (Support Vector Machine) and Nu-SVM. Busulfan is the drug used in this study, and its solubility is measured using a supercritical method and gravimetric technique. The solubility is measured over a wide pressure range of 120–400 bar and at various temperatures in the supercritical range of the used solvent which is CO2 in this case. Because of its advantageous operational properties, supercritical carbon dioxide is used as a solvent. The simulations were carried out using the support vector machine method. The solubility values of busulfan serve as the output, and the pressure and temperature serve as the model's inputs. With the R-square criterion, the Nu-SVM model showed an R-square score of 0.993, and the linear SVM had a score of 0.856. Also, with the MAE metric, Nu-SVM has an error rate of 1.44531E-05, and the linear model has a rate of 7.34302E-05. In addition, according to the MAPE standard, the error of these two models are 1.23755E-01 and 5.12260E-01, respectively.