Advanced simulation and numerical evaluation of pharmaceutical solubility estimation under supercritical processing...

Area of Science:

• Pharmaceutical Science
• Chemical Engineering
• Computational Chemistry

Background:

• Poor aqueous solubility is a major challenge in drug development, limiting bioavailability.
• Supercritical fluid technology, particularly using supercritical carbon dioxide (SC-CO2), offers a promising route to enhance drug solubility.
• Accurate prediction of solubility under varying conditions is crucial for process design and optimization.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting pharmaceutical solubility in SC-CO2.
• To investigate the prediction of SC-CO2 solvent density alongside drug solubility.
• To establish a data-driven framework for supercritical pharmaceutical process design.

Main Methods:

• Utilized a dataset of temperature, pressure, SC-CO2 density, and glibenclamide solubility.
• Employed machine learning models: Convolutional Neural Network (CNN), Multilayer Perceptron (MLP), and Deep Neural Network (DNN).
• Optimized model hyperparameters using the Dragonfly Algorithm (DA).

Main Results:

• MLP model achieved R-squared scores of 0.99093 for solubility and 0.99911 for SC-CO2 density prediction.
• CNN and DNN models also demonstrated high predictive accuracy for both solubility and density.
• Models effectively captured the complex relationships between processing conditions and solubility.

Conclusions:

• Data-driven machine learning models reliably predict drug solubility in SC-CO2.
• These models provide a practical tool for optimizing supercritical processes in pharmaceutical formulation.
• Computational prediction accelerates the design and efficiency of solubility enhancement techniques.