Applicability Domain: A Step Toward Confident Predictions and Decidability for QSAR Modeling

Area of Science:

• Computational toxicology
• cheminformatics
• QSAR modeling

Background:

• The Organization of Economic Co-operation and Development (OECD) mandates a defined applicability domain (AD) for validating quantitative structure-activity relationship (QSAR) models.
• AD estimation is vital for assessing prediction uncertainty by comparing new molecules to training data.
• Characterizing interpolation space is key to defining the AD, an active research area with diverse methodologies.

Purpose of the Study:

• To summarize key concepts of applicability domain (AD) in QSAR modeling.
• To detail various methods for computing AD, including their thresholds and criteria.
• To discuss transparent and decision domains and provide practical guidance for users.

Main Methods:

• Review and synthesis of existing AD methodologies.
• Explanation of training set interpolation in descriptor space.
• Discussion of transparent and decision domain concepts.

Main Results:

• A comprehensive overview of AD concepts and their importance in QSAR.
• Detailed explanation of diverse AD calculation methods and their underlying algorithms.
• Identification of challenges in comparing different AD methods due to methodological variety.

Conclusions:

• Understanding and defining the AD is essential for robust QSAR model validation and reliable safety assessments.
• The multiplicity of AD methods necessitates clear guidance for end-users to ensure appropriate model application.
• Practical examples and open-source tools are provided to facilitate AD determination in QSAR projects.