Machine learning analysis of molecular dynamics properties influencing drug solubility

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Machine learning in pharmacology

Background:

• Aqueous solubility is a critical factor influencing drug bioavailability and therapeutic success.
• Early-stage prediction of solubility minimizes resource waste and enhances clinical trial success rates.
• Molecular dynamics (MD) simulations provide detailed molecular insights into solubility-determining factors.

Purpose of the Study:

• To statistically evaluate the impact of MD-derived properties and logP on drug aqueous solubility.
• To identify key molecular descriptors that significantly influence solubility using machine learning (ML).
• To develop accurate predictive models for aqueous solubility in drug discovery.

Main Methods:

• Compiled a dataset of 211 diverse drugs from existing literature.
• Performed MD simulations to extract ten relevant molecular properties.
• Integrated logP values and selected significant MD features for ML analysis.
• Utilized four ensemble ML algorithms (Random Forest, Extra Trees, XGBoost, Gradient Boosting) for prediction.

Main Results:

• Identified seven key properties (logP, SASA, Coulombic_t, LJ, DGSolv, RMSD, AvgShell) as highly influential for solubility prediction.
• ML models demonstrated predictive performance comparable to structure-based methods.
• Gradient Boosting achieved the highest accuracy with R² of 0.87 and RMSE of 0.537 on the test set.

Conclusions:

• Integrating MD simulations with ML significantly enhances the accuracy and efficiency of aqueous solubility predictions.
• This approach aids in prioritizing drug candidates with optimal solubility early in the development pipeline.
• The study highlights the potential of computational methods to streamline drug discovery and development processes.