Rational approach to the evaluation of short-term tests: analysis of a homogeneous data base.

The major thrust in this study is the analysis by mathematical classification techniques of a homogeneous set of data on short-term tests. The data base generated by the International Program for the Evaluation of Short-Term Tests for Carcinogens was chosen as the model system and was studied first by cluster analysis. For this purpose, an adequate subset of data was selected and the associations between tests were explored on the basis of their performances over a range of chemicals. Then the complementary analysis of the functional relationships between the chemicals, in respect to their genotoxic effects, was performed. In both cases, a clearly structured underlying typology was brought out. Specific indications were suggested by this analysis. First, the data base studied indicated a relative independence of the performances of the assays from phylogeny and genetic endpoints. Second, the chemicals assayed were grouped together (on the basis of functional similarities) according to relationships that may represent a ground for structure-activity relationship investigations. A further important topic of this study was the quantitative evaluation of the test performances in respect to carcinogenicity prediction. To assess the effectiveness with which short-term assays can detect carcinogens and noncarcinogens, discriminant-analysis techniques were applied. On the data set considered here, the short-term tests showed 83.9% global ability to correctly classify chemicals as carcinogens or noncarcinogens. Even if these results are directly dependent on the representativity of the data set chosen, they clearly point out to the high heuristic potency of applying data-analysis techniques to genetic toxicology research.