Applicability domain: towards a more formal definition

Area of Science:

• Computational chemistry
• Toxicology
• Cheminformatics

Background:

• Applicability domain (AD) is vital in quantitative structure-activity relationship (QSAR) modeling for human safety assessment.
• Current AD methods lack a unified definition, causing confusion and hindering model comparison.
• The diversity of AD techniques complicates assessing model adequacy and prediction confidence.

Purpose of the Study:

• To deconstruct the monolithic concept of Applicability Domain (AD) into distinct sub-domains.
• To clarify, formalize, and extend the definition of AD.
• To propose a framework for a unified AD definition in the QSAR community.

Main Methods:

• Conceptual analysis of existing AD methodologies.
• Decomposition of AD into model, prediction, and decision levels.
• Development of a novel framework for AD assessment.

Main Results:

• Demonstrated that AD is not a singular concept but comprises three distinct sub-domains.
• Established a framework to assess confidence at the model, prediction, and decision levels.
• Provided a foundation for a standardized AD definition.

Conclusions:

• A clear, formalized, and extended definition of AD is achievable through its sub-domain decomposition.
• The proposed framework facilitates a common understanding and comparison of AD across different QSAR models.
• Encourages community-wide convergence towards a unified AD definition for enhanced reliability in safety assessments.