Edward Thompson

The use of substructure search and relational databases for examining the carcinogenic potential of chemicals.

The Procter and Gamble Co., Miami Valley Labs, PO Box 398707, Cincinnati, OH 45239-8707, USA. Tel: 513-627-2245, Fax: 513-627-1908, E-mail: Thompson.ed@pg.com

Attempts to predict biological endpoints based on chemical structure and/or properties have produced limited successes. The accuracy of these predictions tends to decline as the chemical structures become more diverse and the endpoints more complex. Those models that do accurately predict biological endpoints have generally been developed with congeneric data sets. In order to use congeneric data sets for our analyses we have built a database of over 52,000 genetox and carcinogenicity test results on 13,000 chemicals. The data are stored in a relational database (ORACLE) and is accessed by substructure search. The data were gathered from publications from the U. S. EPA, FDA, NCI, The National Toxicology Program, IARC, RTECS, EMIC, The Gold Carcinogenicity database, and the open literature. Two features of this database make it unique. First, substructure search allows us to find data for structural analogs of the chemical of interest. The identification of structural analogs provides us valuable information for making decisions when no data exist for the chemical of interest. These decisions are sometimes qualitative (a simple visual inspection of the data), or quantitative (a QSAR model would be developed). Whatever the process, the decision will be based on data from closely related chemical structures. Second, storing the data in a relational database allows us to relate biological effects to chemical structures, e.g., it is a simple matter to determine which of the over 4,000 chemicals with rodent bioassay data produce carcinomas in the liver of male rats. Since this is a relatively specific biological effect, one might expect those chemicals to be structurally related. This, of course is not the case. The group of chemicals which produce carcinomas in liver of male rats is structurally diverse. This and other examples which demonstrate the unique power that relational databases provide will be discussed. Lastly, an in depth look at the carcinogenic potential of azo dyes will be presented. The presentation will demonstrate how a consumer products company would use a database of this type to guide product development. It will also demonstrate potential pitfalls of applying QSAR tools without a thorough understanding of the biology.

Back to Program Page
Back to Main Page