Development and prevalidation of a list of structure-activity relationship rules to be used in expert systems for prediction of the skin-sensitising properties of chemicals.

The new European Union (EU) chemicals policy, as described in the White Paper entitled Strategy for a Future Chemicals Policy, has identified a need for computer-based tools suitable for predicting the hazardous properties of chemicals. Two sets of structural alerts (fragments of chemical structure) for the prediction of skin sensitisation hazard classification ("R43, may cause sensitisation by skin contact") have been drawn up, based on sensitising chemicals from a regulatory database (containing data for the EU notification of new chemicals). These alerts comprise 15 rules for chemical structures deemed to be sensitising by direct action of the chemicals with cells or proteins within the skin, and three rules for substructures that act indirectly, i.e. requiring chemical or biochemical transformation. The predictivity rates of the rules were found to be good (positive predictivity, 88%; false-positive rate, 1%; specificity, 99%; negative predictivity, 74%; false-negative rate, 80%; sensitivity, 20%). Because of the confidential nature of the regulatory database, the rules are supported by examples of sensitising chemicals taken from the "Allergenliste" now held by the Federal Institute for Risk Assessment (BfR) and the DEREK for Windows expert system. The rules were prevalidated against data not used for their development. As a result of the prevalidation study, it is proposed that the two sets of structural alerts should be taken forward for formal validation, with a view to incorporating them into regulatory guidelines.