Predictive QSAR modeling based on diversity sampling of experimental datasets for the training and test set selection

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Predictive power is crucial for Quantitative Structure Activity Relationship (QSAR) models.
• QSAR models predict biological activity for novel compounds.
• Accurate prediction requires robust model validation using distinct training and test sets.

Purpose of the Study:

• To enhance QSAR model predictive power through rational dataset division.
• To establish criteria for selecting training and test sets based on molecular descriptor space.
• To introduce and apply diversity indices and sphere-exclusion algorithms for optimal set partitioning.

Main Methods:

• Utilizing molecular dataset diversity indices for quantitative criteria.
• Employing sphere-exclusion algorithms for rational training and test set selection.
• Validating QSAR models built with the proposed method on experimental datasets.

Main Results:

• QSAR models developed using rational set division demonstrate superior predictive power.
• The proposed method significantly outperforms random or activity ranking based selection.
• Diversity principles ensure representative and robust training and test sets.

Conclusions:

• Rational selection of training and test sets based on diversity is essential for QSAR modeling.
• This approach improves the reliability and accuracy of QSAR predictions.
• Routine application of these methods will advance QSAR research and drug discovery.