Supervised self-organizing maps in drug discovery. 2. Improvements in descriptor selection and model validation

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Modeling nonlinear descriptor-target relationships is crucial in drug discovery.
• Self-organizing maps (SOMs) are promising for decoding these complex relationships.
• Traditional methods may not fully capture predictive power.

Purpose of the Study:

• To develop high-quality, predictive quantitative structure-activity/property relationship (QSAR/QSPR) models.
• To combine supervised self-organizing maps with simulated annealing for enhanced modeling.
• To validate model performance using external datasets and novel techniques.

Main Methods:

• Utilized a supervised self-organizing map (SOM) combined with simulated annealing (SA).
• Applied the technique to six diverse biological endpoint datasets.
• Employed external validation, cross-validation, and dynamic partitioning for robust assessment.

Main Results:

• Successfully built highly predictive QSAR/QSPR models across various biological endpoints.
• Demonstrated the efficacy of the SOM-SA combination for complex relationship modeling.
• Confirmed model quality through rigorous external validation, mitigating overfitting.

Conclusions:

• The integrated SOM-SA approach provides a powerful tool for drug discovery.
• External validation and advanced cross-validation are essential for assessing predictive accuracy.
• This method advances the development of reliable structure-activity/property models.