Practical explanation of applicability domain in regulatory (Q)SAR and why domain assessment is model-specific.
Applicability domain defines where a model has sufficient structural, physicochemical and response-space support for its prediction. It is a central part of defensible regulatory (Q)SAR interpretation.
Model-specific concept
A rule-based system and a similarity-driven statistical model do not necessarily define domain in the same way.
Out-of-domain does not equal positive or negative
It indicates reduced support for the prediction, not a hazard conclusion by itself.
Similarity and representation
Statistical models often rely on chemical similarity or descriptor-space coverage to evaluate domain.
Expert response
An out-of-domain result may require additional models, analogues, experimental evidence or a more cautious conclusion.
Frequently asked questions
Is applicability domain the same as confidence?
No. They are related but distinct concepts.
Can an out-of-domain prediction still be informative?
Yes, but with recognized limitations.
Should every QMRF describe applicability domain?
OECD principles expect a defined applicability domain.
Can different models give different domain conclusions?
Yes, because their representations and training spaces differ.
