POSTER ABSTRACT / DETAILS:
Read-across is a powerful approach to address data gaps for REACH. The most important issues that influence uncertainty in toxicity prediction are the experimental data used for data gap filling and the weight of evidence supporting the categories formation approach. Read - across requires good quality experimental in vivo data for the regulatory endpoint of concern. Therefore, a clear and detailed explanation of the origin of the data is needed for an assessing the uncertainty associated with them. The grouping schemes and profilers should clearly group chemicals into a toxicologically meaningful category supported by scientific evidence and explanation (i.e. mechanistic transparency).
This work addresses the issues of reliability and relevance of constitutive elements of the OECD QSAR Toolbox 3.2 software, namely elements of data gap filling workflow: databases and profilers using for category formation and read-across. The case studies selected in this work are relative to genotoxicity and aquatic toxicity.
The databases in the Toolbox software were not created ad hoc; but have been donated by different bodies, without selection or curation of data by the Toolbox developers. This implies that the characteristics and quality of different databases are remarkably different. We have examined a number of different possibilities for parameterizing reliability both for the entire databases and for individual datapoints. The reliability scores for six database for aquatic toxicity and eight databases for genotoxicity have been estimated. The usefulness of the data reliability score for improving of read-across confidence has been demonstrated in practical case studies.).
For the profilers, it has been shown that positive predictivity is a good reliability measure for some of them (e.g., Ames mutagenicity), whereas it does not apply well to profilers for aquatic toxicity, that are aimed at building of toxicologically plausible mechanistic classes, instead of predicting directly a toxicological endpoint. For profiler relevant to the aquatic toxicity endpoint the goodness-of-fit of regression model has been used as a measure for reliability scores. In this work, the reliability for five different profilers for aquatic toxicity and seven profilers for genotoxicity have been estimated.
In conclusion, Read-across is not a trivial undertaking. Its successful applications strongly depend on a range of elements such as the availability of datasets and profilers, together with mechanistic transparency, information on the level of reliability of each of them and relevance of such elements to the endpoint of interest.
This work was supported by Service Contract ECHA/2013/167 “Scientific Review of the QSAR Toolbox and usability improvements”