Title: Collection and Evaluation of (Q)SAR Models for Mutagenicity and Carcinogenicity
Authors: NETZEVA TATIANAWORTH ANDREWBENIGNI RomualdoBOSSA Cecilia
Publication Year: 2007
JRC N°: JRC36970
Other Identifiers: EUR 22772 EN
URI: http://publications.jrc.ec.europa.eu/repository/handle/JRC36970
Type: EUR - Scientific and Technical Research Reports
Abstract: This evaluation of the non-commercial (Q)SARs for mutagenicity and carcinogenicity consisted of a preliminary survey (Phase I), and then of a more detailed analysis of short listed models (Phase II). In Phase I, the models were collected from the literature, and then assessed according to the OECD principles based on the information provided by the authors-. Phase I provided the support for short listing a number of promising models, that were analyzed more in depth in Phase II. In Phase II, the information provided by the authors was completed and complemented with a series of analyses aimed at generating an overall profile of each of the short listed models. The models can be divided into two families based on their target: a) congeneric; and b) non-congeneric sets of chemicals. The QSARs for congeneric chemicals include most of the chemical classes top ranking in the EU High Production Volume list, with the notable exception of the halogenated aliphatics. They almost exclusively aim at modeling Salmonella mutagenicity and rodent carcinogenicity, which are crucial toxicological endpoints in the regulatory context. The lack of models for in vivo genotoxicity should be remarked. Overall the short listed models can be interpreted mechanistically, and agree with, and/or support the available scientific knowledge, and most of the models have good statistics. Based on external prediction tests, the QSARs for the potency of congeneric chemicals are 30 to 70 % correct, whereas the models for discriminating between active and inactive chemicals have considerably higher accuracy (63 to 100 %), thus indicating that predicting intervals is more reliable than predicting individual data points. The internal validation procedures (e.g., cross-validation, etc...) did not seem to be a reliable measure of external predictivity. Among the non-local, or global approaches for non-congeneric data sets, four models based on the use of Structural Alerts (SA) were short listed and investigated in more depth. The four sets did not differ to a large extent in their performance. In the general databases of chemicals the SAs appear to agree around 65% with rodent carcinogenicity data, and 75% with Salmonella mutagenicity data. The SAs based models do not seem to work equally efficiently in the discrimination between active and inactive chemicals within individual chemical classes. Thus, their main role is that of preliminary, or large-scale screenings. A priority for future research on the SAs is their expansion to include alerts for nongenotoxic carcinogens. A general indication of this study, valid for both congeneric and noncongeneric models, is that there is uncertainty associated with (Q)SARs; the level of uncertainty has to be considered when using (Q)SAR in a regulatory context. However, (Q)SARs are not meant to be black-box machines for predictions, but have a much larger scope including organization and rationalization of data, contribution to highlight mechanisms of action, complementation of other data from different sources (e.g., experiments). Using only non-testing methods, the larger the evidence from QSARs (several different models, if available) and other approaches (e.g. chemical categories, read across) the higher the confidence in the prediction.
JRC Institute:Institute for Health and Consumer Protection

Files in This Item:
File Description SizeFormat 
6970 - EUR 22772 - EUR_CM_31May07.pdf1.38 MBAdobe PDFView/Open


Items in repository are protected by copyright, with all rights reserved, unless otherwise indicated.