Predictions based upon the CASE structure-activity relational method are independent of the nature of the chemicals in the data base.

It is demonstrated that the predictivity of the CASE structure-activity relational method is determined by the number of chemicals in the learning set but not by the nature of the chemicals. Thus learning sets containing equal numbers of different chemicals will be equally predictive of the activity of chemicals not in the learning sets. These findings have implications with respect to the strategy of selecting chemicals for listing and inclusion in data bases as well as for mechanistic studies in which different toxicological endpoints are compared.