Systematic Modeling of log D_7.4 Based on Ensemble Machine Learning, Group Contribution, and Matched Molecular Pair...

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Lipophilicity, measured as the distribution coefficient (log D7.4), is crucial for predicting drug candidate ADMET properties.
• Accurate prediction of lipophilicity is essential for efficient drug discovery and development.

Purpose of the Study:

• To develop robust quantitative structure-property relationship (QSPR) models for predicting log D7.4.
• To compare the performance of various machine learning algorithms and modeling approaches.

Main Methods:

• Utilized a large, diverse dataset of drug candidates.
• Employed eight machine learning algorithms with 43 selected molecular descriptors.
• Developed a consensus model from the top three performing algorithms.
• Validated models using Y-randomization and applicability domain analysis.

Main Results:

• The XGBoost model showed strong prediction performance (RT² = 0.906).
• The consensus model further improved prediction accuracy (RT² = 0.922, RMSEt = 0.359).
• The descriptor-based consensus model outperformed group contribution and local models.

Conclusions:

• The developed consensus QSPR model is a reliable and promising tool for evaluating log D7.4 in drug discovery.
• The study provides insights into structure-property relationships influencing lipophilicity.