Solubility of Pharmaceutical Compounds in Organic Solvents Using Artificial Nural Network and Correlation Model

In their earlier research, the authors explored the correlation of solid-liquid solubility using an updated semi-theoretical equation (Eur. J. Pharm. Sci. 2019, 143, 1-13). This current study applies an Artificial Neural Network (ANN) and a refined Apelblat model to predict the solubility of nine pharmaceutical compounds in pure organic solvents over an extensive temperature range. By training the optimized network using the back-propagation method of the Levenberg-Marquardt algorithm, optimal parameters —such as neurons, hidden layers, and transfer function — were established in the ANN through the utilization of training, testing, and validation data. This network was trained with 764 data points and subsequently tested and validated with 164 data points. A satisfactory correlation of 1.33% was obtained from the enhanced Apelblat model across 1092 data points. The findings from thermodynamic analysis and solubility parameters can optimize the purification process in the synthesis of pharmaceutical compounds.