The broad, long-term objectives of this project are to:

1. Develop predictive toxicology methods that can help prioritize the selection of test agents for expensive animal testing, and identify specific testing needs for each chemical;
2. Help reduce the utilization of laboratory animals, maximize the information derived from testing that is conducted, and manage limited resources most effectively;
3. Assist the science of toxicology to progress beyond reliance on empirical testing and statistical extrapolation for risk assessment of individual chemicals, into a maturing science with an expanded base of rules, hypotheses, mechanistic inference, deductive experimentation, modeling, and theory validation.

The specific aims of this project are to:

1. Determine for noncongeneric chemical test agents the features of these agents and of biological host systems, and the relationships among these features, which give rise to toxic effect, by applying new intelligent computer system techniques to existing animal bioassay and physico-chemical structure data;
2. Develop several distinct and independent models with predictive value for toxic effect and human hazard identification, utilizing methods that will yield useful information and insights into the operative mechanistic pathways;
3. Perform a series of evaluative experiments using these models. These experiments include investigating the impact of choice of chemical information and toxicological domain expertise on efficiency, accuracy, and comprehensibility of these models;
4. Investigate methods of combining various predictive models so as to leverage strengths while compensating for weaknesses in individual models, employing combination methods which apply equally well to predictive models regardless of their characteristics.

Research design falls into three complementary thrusts:

1. Development of models;
2. Use of models;
3. Combination of models.