Subhash C. Basak, Gregory D. Grunwald and Brian D. Gute
Use of computational methods in predicting potential toxicity of
Research Institute, University of Minnesota-Duluth, Duluth, MN 55811, USA
A recent trend in predictive toxicology is the prediction of potential toxicity of chemicals and xenobiotics using computational methods. There are thousands, perhaps millions, of chemicals which are used for various industrial, cosmetic, and medicinal purposes and human beings are exposed to a large number of them. Also, the chemicals may be released into the environment. All these chemicals may adversely affect human and environmental health.
US Air Force routinely uses many chemicals some of which may be hazardous to human health and the environment. Therefore, it is essential that proper risk assessment of these chemicals is carried out. A large number of experimental data are necessary for such hazard/ risk assessment of chemicals in an effective manner. But the scarcity of resources and testing facilities does not allow us to test all such chemicals exhaustively in the laboratory. One viable alternative in this situation is the use of computational methods for estimating potential toxicity of chemicals. These techniques may help us in prioritizing chemicals based on their potential (computed) toxicity.
Our research group has been involved in the development of the theoretical basis and methods for prediction of hazardous potential of chemicals. We have been involved primarily in three different areas: a) hierarchical quantitative structure-activity relationships (QSARs), b) quantitative molecular similarity analysis (QMSA), and c) prediction of modes of action (MOA) of chemicals from structure.
In hierarchical QSAR, one uses primarily computed parameters, viz., topostructural, topochemical, geometrical, and quantum chemical indices, in a hierarchical manner. In this approach, more complex parameters are used for model building, if necessary. If computed parameters are not sufficient to develop good models, simple experimental properties may be used for QSAR development.
In QMSA, computed structural indices are used to define intermolecular similarity and intermolecular distance in structure spaces. The computed similarity scores are then used to select analogs. Properties of selected analogs are then used to estimate properties of query chemicals based on K-nearest neighbor (KNN) method. Such estimated properties can be use when good QSARs are not available for some chemical.
Generally, it is better to use a QSAR model for a specific class of compound (with specific MOA) than using any arbitrary QSAR equation blindly. This involves a two-tier approach: a) first predict the MOA of the chemical from structure, and b) then predict the potential toxicity of the chemical using the appropriate class-specific model.
Utility of the three computational approaches in predictive toxicology and hazard assessment of chemicals will be discussed.Back to Program Page