Advanced hybrid computational analysis of febuxostat solubility using machine learning in supercritical processing...

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Materials Science

Background:

• Supercritical fluids (SCFs) are increasingly used as sustainable alternatives to organic solvents in industrial applications.
• SCFs, particularly supercritical CO2, show significant potential for enhancing the solubility of poorly water-soluble drugs.
• SCF-based processes offer advantages such as eco-friendliness, cost-effectiveness, safety, and improved product purity compared to traditional methods.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the solubility of febuxostat (FBX) in supercritical fluids.
• To investigate the influence of temperature and pressure on FBX solubility using computational modeling.
• To compare the performance of different regression models and optimization techniques for solubility prediction.

Main Methods:

• Utilized machine learning regression models: Gaussian Process Regression (GPR) and K-Nearest Neighbors (KNN).
• Developed a voting regression model combining GPR and KNN predictions.
• Employed the Harris Hawks Optimization (HHO) algorithm for hyper-parameter tuning of the machine learning models.

Main Results:

• The GPR model achieved an R² score of 0.819, and the KNN model achieved 0.854.
• The voting regression model demonstrated superior performance with an R² score of 0.980.
• The optimized voting model exhibited low error rates: RMSE of 2.78 × 10⁻¹ and MAPE of 3.81 × 10⁻².

Conclusions:

• The combined voting regression model significantly outperforms individual GPR and KNN models in predicting febuxostat solubility.
• Hyper-parameter optimization using the HHO algorithm enhances the predictive accuracy of the solubility models.
• Machine learning approaches, particularly the optimized voting model, provide a reliable and efficient method for modeling drug solubility in supercritical fluids.