Predicting drug solubility in supercritical carbon dioxide green solvent using machine learning models based on...

Area of Science:

• Pharmaceutical Science
• Chemical Engineering
• Computational Chemistry

Background:

• Accurate drug solubility prediction in supercritical carbon dioxide (scCO₂) is vital for pharmaceutical processes like particle engineering and extraction.
• Experimental solubility determination is resource-intensive, necessitating advanced predictive methods.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting drug solubility in scCO₂.
• To identify the most reliable and accurate model for solubility prediction beyond the training data range.

Main Methods:

• Utilized machine learning algorithms including CatBoost, XGBoost, LightGBM, and Random Forest (RF).
• Trained and tested models on solubility data for sixty-eight drugs in scCO₂.
• Performed statistical error analysis and graphical assessments to compare model performance.

Main Results:

• The XGBoost model exhibited the highest accuracy and reliability, outperforming other evaluated models.
• XGBoost achieved a low root mean square error (RMSE) of 0.0605 and a high R² value of 0.9984.
• 97.68% of data points fell within the XGBoost model's applicability domain, confirming its predictive robustness.

Conclusions:

• The XGBoost algorithm provides a robust and efficient method for estimating drug solubility in scCO₂.
• Machine learning models, particularly XGBoost, offer a significant advantage over traditional methods for solubility prediction.
• This approach facilitates efficient pharmaceutical process design and development.