Evaluating batteries of short-term genotoxicity tests

Publisher Website
Google Scholar
Abstract
Selecting a battery of short-term genotoxicity tests suitable for screening unknown chemicals for carcinogenicity can be a large combinatorial task because of the great number of short-term tests currently available. Biological criteria, such as requirements for different targets and endpoints, can reduce the number of possible combinations. An independent yet potentially complementary approach which we have developed uses Bayes' theorem to predict carcinogenicity from results (positive or negative) in short-term tests. Batteries can be evaluated by their predictivity, calculated with Bayes' theorem from the sensitivities and specificities of the component tests. Our analyses indicate that tests which contribute most to a battery's predictivity are those which are both sensitive and specific, which we call Class I tests. Because few of the currently available tests are Class I, we have extended our analyses to consider when other types of tests must be substituted for Class I tests, the purpose of a particular test program will influence the choice. In order to obtain the best predictions for all chemicals, a battery should include an equal number of Class II tests (i.e. those that are sensitive but are not specific) and Class III tests (i.e. those that are not sensitive but are specific). However, for the purpose of reducing the number of carcinogens erroneously classified as noncarcinogenic, Class II tests contribute about twice as much to the predicitivity of a battery as do Class III tests, and for the purpose of reducing the number of non-carcinogens erroneously classified as carcinogenic, Class II tests contribute about half as much as Class III tests. (Exact equivalences of tests depend upon the values of sensitivity and specificity.) We have also developed a graphical representation of the predictivity of batteries, which takes into consideration the frequency of occurrence of a set of results, which is easily calculated from their sensitivities and specificities. These twodimensional representations allow the evaluation of batteries based on a number of different criteria. The Bayesian approach can be used in conjunction with other methods of test selection to construct batteries which satisfy biological criteria and which are also highly predictive.

This publication has 2 references indexed in Scilit: