A Distance-Based Boolean Applicability Domain for Classification of High Throughput Screening Data.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Quantitative Structure-Activity Relationship (QSAR) modeling requires both activity prediction models and applicability domains.
• Existing applicability domain methods can be complex, difficult to implement, and hard to interpret, often requiring threshold selection or empirical constants.

Purpose of the Study:

• To introduce a fully automatic and easily interpretable Distance-Based Boolean Applicability Domain (DBBAD) algorithm for category QSAR.
• To evaluate the effectiveness of the DBBAD algorithm in improving QSAR model performance and applicability domain definition.

Main Methods:

• Development of the Distance-Based Boolean Applicability Domain (DBBAD) algorithm for category QSAR.
• Retrospective experimental validation using High Throughput Screening (HTS) datasets.
• Integration and comparison with support vector machine (SVM) and random forest (RF) classifiers.

Main Results:

• The DBBAD algorithm demonstrated improved classification performance for SVM and RF classifiers.
• Early retrieval of active molecules was enhanced by the DBBAD applicability domain.
• Improved scaffold diversity was observed among the top-ranked active molecules identified.

Conclusions:

• The DBBAD algorithm provides a simple, automatic, and interpretable solution for defining applicability domains in QSAR.
• This method enhances the reliability and efficiency of QSAR models, particularly for category QSAR applications.
• The DBBAD approach contributes to better hit identification and scaffold diversity in drug discovery efforts.