Utilization of advanced machine learning models for analysis of pharmaceutical cocrystals by prediction of solubility...

Area of Science:

• Materials Science
• Computational Chemistry
• Pharmaceutical Sciences

Background:

• Pharmaceutical cocrystals improve solubility and bioavailability of poorly water-soluble drugs.
• Coformer selection is critical for successful cocrystal design, requiring understanding of molecular interactions.
• Predicting coformer properties is essential for efficient pharmaceutical cocrystal development.

Purpose of the Study:

• To develop robust machine learning models for predicting Hansen solubility parameters (HSPs) of pharmaceutical cocrystals.
• To utilize molecular descriptors from COSMO-RS and group contribution methods for model development.
• To identify the most effective machine learning model for HSP prediction in cocrystal screening.

Main Methods:

• A dataset of 181 samples with 86 features, including COSMO-RS and group contribution descriptors, was used.
• Data preprocessing involved outlier detection (Isolation Forest) and feature selection (SFFS).
• Random Forest, Extra Trees, and Gradient Boosting Regression Trees models were developed and optimized using the Dragonfly algorithm.

Main Results:

• The Extra Trees (ET) model consistently outperformed Random Forest and Gradient Boosting Regression Trees models.
• The ET model achieved high test-set R² values (0.9175, 0.8661, 0.9815) for the three HSPs.
• Feature importance analysis indicated significant contributions from both group contribution and COSMO-RS descriptors.

Conclusions:

• The developed machine learning framework offers an efficient and accurate method for predicting HSPs and screening cocrystals.
• The Extra Trees model shows superior predictive capability for coformer selection in pharmaceutical cocrystal design.
• Combining molecular thermodynamic descriptors with machine learning enhances the design of pharmaceutical cocrystals.