Three useful dimensions for domain applicability in QSAR models using random forest.

Area of Science:

• Computational Chemistry
• cheminformatics
• Drug Discovery

Background:

• Quantitative structure-activity relationship (QSAR) models are crucial for predicting compound activity.
• Traditional QSAR accuracy estimation relies heavily on compound similarity to training sets.
• Emerging research suggests similarity alone may be insufficient for robust accuracy assessment.

Purpose of the Study:

• To develop a more accurate method for estimating QSAR prediction accuracy.
• To incorporate additional parameters beyond compound similarity into accuracy estimation.
• To validate the predictive power of the proposed accuracy estimation method.

Main Methods:

• Utilized random forest tree variation and prediction range as additional accuracy metrics.
• Developed a three-dimensional array of bins for storing cross-validation results.
• Applied the method to ensemble-based QSAR models.

Main Results:

• Simultaneous use of similarity, tree variation, and prediction range significantly improved prediction accuracy discrimination.
• Cross-validation root-mean-square errors accurately predicted errors for external test compounds.
• The proposed method enhances the reliability of QSAR model performance evaluation.

Conclusions:

• A novel, multi-parameter approach offers superior QSAR prediction accuracy estimation.
• This method is applicable to ensemble QSAR modeling techniques.
• The findings improve the reliability of prospective QSAR predictions in drug discovery.