Automated machine-learning framework for predicting drug solubility in supercritical CO₂ for sustainable process...

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Green Chemistry

Background:

• Experimental drug solubility measurements in supercritical carbon dioxide (SC-CO2) are slow and resource-intensive, hindering green pharmaceutical process development.
• Existing methods limit the advancement of eco-friendly technologies like particle formation and solvent-free formulations.

Purpose of the Study:

• To develop an automated computational framework for accurate and scalable prediction of drug solubility in SC-CO2.
• To provide a practical alternative to extensive laboratory experiments for pharmaceutical process design.

Main Methods:

• Coupling Adaptive Boosting Regression and Light Gradient Boosting Regression with bio-inspired optimization (Osprey Optimization Algorithm, Artificial Protozoa Optimizer).
• Utilizing hybrid ensemble schemes and metaheuristic algorithms for model tuning.
• Assessing model performance through cross-validation, accuracy metrics (RMSE, R2, MDAPE), statistical comparison, and sensitivity analysis.

Main Results:

• The Artificial Protozoa Optimizer-driven ensemble (ALAP) demonstrated superior performance.
• Achieved high accuracy with RMSE = 0.191, R2 = 0.982, and MDAPE = 15.6% on the test set.
• Identified ALAP as the most reliable configuration through multi-criteria ranking (TOPSIS).

Conclusions:

• The developed computational framework offers a reliable and efficient tool for predicting drug solubility in SC-CO2.
• This data-driven approach supports the design of environmentally conscious pharmaceutical processes.
• The framework serves as a practical, transferable alternative to traditional experimental methods.