Computer Assisted Methods In Predictive Toxicology

The list of chemicals for which no toxicity data exist is ballooning. This requires both federal regulatory agencies and industry to explore new rapid and cost-effective methods for assessing health hazards and environmental effects of chemicals, especially since the resources are scarce and risk assessment is becoming a social priority. The research is directed at the development of computer-assisted methods for predictive toxicology. Specifically, during Phase I, cross-validated structure-toxicity models will be developed for confident prediction of the carcinogenic potential of untested chemicals. In order to better understand the models and predictions therefrom, not only the transport, electronic, and steric attributes of molecular structure, but also the results form certain short-term toxicological studies along with organ-specific toxicity data will be used as descriptor variables. Both parametric statistical and supervised artificial intelligence techniques will be employed to explore non-linear relationships between these descriptors and as complex a phenomenon as carcinogenesis. The success of Phase I will provide impetus to (1) extend the technique to other neoplastic and non-neoplastic toxic endpoints, and (2) automate the model building and model installation techniques. This would result in an artificial intelligence package for predicting a wide toxicity profile of untested chemicals.