Development of advanced model for understanding the behavior of drug solubility in green solvents: Machine learning modeling for small-molecule API solubility prediction

Determination of small-molecule API (Active Pharmaceutical Ingredient) solubility in solvents is of great importance for drug development in pharmaceutical industry. This study uses machine learning algorithms to predict the solubility of phenytoin drug and density of the solvent at different pressure and temperature. In fact, the solubility of drug (y) and the density of supercritical CO2 (Sc-CO2) were investigated as functions of temperature (T) and pressure (P). Three regression models, namely Gaussian Process Regression (GPR), Gamma Regression (GR), and K-nearest neighbor (KNN) were employed to predict the output variables. To optimize the performance of these models, the Directional Bat Algorithm (dBA) was utilized for hyper-parameter tuning. The outputs demonstrated the accuracy of the developed models in predicting the Sc-CO2 density and solubility of phenytoin. Among the models, GPR exhibited the highest accuracy for both outputs, achieving an R2 value of 0.99 for the solvent density and 0.999 for the solubility. The root mean square error (RMSE) for Sc-CO2 density was approximately 18.37, with an average absolute relative deviation percentage (AARD%) of 4.07% and a maximum error of 37.42. For solubility, the RMSE was approximately 0.087, with an AARD% of 2.29% and a maximum error of 0.150.