Structure activity-based predictive toxicology: an efficient and economical method for generating non-congeneric data bases.

Starting with the results of the mutagenicity of 100 chemicals, we use CASE, a structure activity relational system, to select additional chemicals for inclusion into a data base. We demonstrate that a data base consisting of approximately 180 carefully selected chemicals is as predictive as the analysis based upon 820 'unselected' chemicals.

Value-of-information analysis of testing strategies: estimating the effect of uncertainty about the proportion of chemicals that are true human carcinogens.

Choosing a cost-effective strategy for classifying chemicals as human carcinogens and non-carcinogens depends upon the costs of false positives (carcinogens erroneously treated as non-carcinogenic) and false negatives (non-carcinogens erroneously treated as carcinogenic); upon the accuracy (sensitivity and specificity) of the classification strategy; and upon the underlying proportion of carcinogens in the population of chemicals to be classified. If these values are known, value-of-information analyses can indicate the most cost-effective among three strategies: classify as carcinogenic without testing, classify as non-carcinogenic without testing, or choose the most cost-effective test and classify on the basis of the test result. When some or all of the values are uncertain, the analysis becomes more complex, but still helps to guide decisions among the three classification strategies.

Methodologies for interpretation of short-term test results which may allow reduction in the use of animals in carcinogenicity testing.

Assessment of the risk to humans posed by chemical substances currently relies primarily on experimental exposure of animals in lifetime feeding studies. Short-term tests for genotoxicity are much less costly and use fewer or no animals, but have not replaced the long-term animal bioassay because their results do not coincide completely. We have developed methodologies for interpretation of short-term tests which improve the usefulness of their results, and may allow them to replace the long-term animal bioassay in some circumstances.

The influence of the proportion of carcinogens on the cost-effectiveness of short-term tests.

The cost-effectiveness of using short-term genotoxicity tests to screen unknown chemicals for carcinogenicity depends upon the inherent reliability of the tests (sensitivity, or fraction of carcinogens giving positive results, and specificity, or fraction of non-carcinogens giving negative results) and also upon the proportion of carcinogens in the population of chemicals to be screened. Individual tests may be combined into batteries to improve reliability; however, this requires decision rules to declare the overall result positive or negative. A framework for developing such rules based upon minimizing costs of false-positives and false-negatives was presented in a seminal paper by Lave and Omenn (1986, Nature (London), 324, 29-34). We have extended their work, which is based on logit analysis, to consider, using Bayes' theorem, the influence of the proportion of carcinogens upon the decision rules for declaring a battery result positive or negative. If the proportion of carcinogens is high (20% or greater), then the most effective tests are those with high sensitivity, and if the proportion of carcinogens is low, then the most effective tests are those with high specificity.

Evaluation of the genotoxicity of theobromine and caffeine.

Published data on the mutagenicity and genotoxicity of theobromine and caffeine were analysed by the Carcinogen Prediction and Battery Section (CPBS) method. In spite of some positive responses, these analyses did not predict for theobromine a potential for causing cancer by virtue of a genotoxic mechanism. Caffeine, on the other hand, clearly has potential for genotoxic carcinogenicity. The predictive performance of cost-effective batteries consisting of selected combinations of four assays was also evaluated. The predictions were similar to those derived when all the available test results were considered.

Artificial intelligence and Bayesian decision theory in the prediction of chemical carcinogens.

Two procedures for predicting the carcinogenicity of chemicals are described. One of these (CASE) is a self-learning artificial intelligence system that automatically recognizes activating and/or deactivating structural subunits of candidate chemicals and uses this to determine the probability that the test chemical is or is not a carcinogen. If the chemical is predicted to be carcinogen, CASE also projects its probable potency. The second procedure (CPBS) uses Bayesian decision theory to predict the potential carcinogenicity of chemicals based upon the results of batteries of short-term assays. CPBS is useful even if the test results are mixed (i.e. both positive and negative responses are obtained in different genotoxic assays). CPBS can also be used to identify highly predictive as well as cost-effective batteries of assays. For illustrative purposes the ability of CASE and CPBS to predict the carcinogenicity of a carcinogenic and a non-carcinogenic polycyclic aromatic hydrocarbon is shown. The potential for using the two methods in tandem to increase reliability and decrease cost is presented.