Alert performance: A new functionality in the OECD QSAR Toolbox
Computational Toxicology Volume 10 (2019) 26-37
Both industry and regulatory agencies use structure-activity relationships and read-across as toxicity assessment methods alternative to animal experimentation. A prerequisite for acceptance of the outcome of such in silico predictions is establishing the mechanistic similarity (i.e. a common mode of (toxicological) action) among the target substance(s) and the source substance(s). To assist in assessing this similarity, an "Alert performance" functionality has been added to the OECD QSAR Toolbox Version 4.0. This functionality is designed to provide information for evaluating the mechanistic and/or structural similarity among the analogues forming a category. This "Alert performance" informs the user on whether the identified alert(s) in the target substance or its metabolites is linked to consistent effects for the endpoint of interest for other substances triggering the same alert and for which experimental data are available. In this respect, calculation of alert performance shows the predictivity of a given alert to the selected endpoint. The predictivity of alerts is context-dependent and is affected by the structure of target chemical, targeted endpoint, selected databases, data usage and data scale. The use of alerts with high performances will lead to the formation of categories of chemicals showing similar toxicological effects, thus reducing the uncertainty when filling data gaps. This paper illustrates the workflow of the alert performance module and addresses several questions that could arise during its applications.