Warning potentially naïve users around the best usage of a predictive algorithm is a significant concern for developers of regulatory and commercial Quantitative Structure Activity/Property Relationship models. However, the development of a general, consistent applicability domain (AD) remains an unsolved problem within the field, with many competitor methods and theories prescribing differing approaches. This work rigorously considers mathematical argumentation behind several benchmark methods. Subsequent experiments on 12 physicochemically- and toxicologically-relevant endpoints examine if candidate ADs suitably improve model performance or otherwise identify regions of chemical space where model prediction is unreliable. This provides a reusable framework by which to characterize successful AD methodologies and assist future efforts perfecting solutions to this outstanding problem.

Comparative analysis of applicability domain methods

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.