![[logo] US EPA](http://www.epa.gov/epafiles/images/logo_epaseal.gif){kind=logo}
2007 CompTox Forum
Abstract - Understanding Toxicity through Chemical and Biological Fingerprints
Chihae Yang, Ph.D.
Vice President, Toxicology & Predictive Modeling
Leadscope, Inc.
1393 Dublin Road, Columbus, OH 43215
Phone: 614-675-3730
E-mail: cyang@leadscope.com
Understanding toxicity endpoints is a fundamentally important aspect of the risk assessment process. Chemically-induced toxicity involves relating mechanisms in the biological domain with molecular structures in the chemical domain. Traditional structure-activity relationships and chemical structure analog-based methods have been, to some extent, successfully applied to understanding toxicity in certain cases such as genotoxic carcinogens. These methods rely on finding close analogs in chemical structure space, which are then projected onto the biological domain to relate a desired endpoint. Advances in computational toxicology are being driven by expanding capabilities for simultaneously mining biological and chemical domains. Methodologies are being developed to link the two domains by finding biological analogs (i.e., similar chemical structures with biologically similar functions) and profiling the domains to find correlations. One way to expand the structural hypothesis in the chemical domain is to use structural descriptors or features of a compound rather than the compound itself. The data spaces of structure-bioassays and structure-toxicity are very sparse at the compound level, and so expansion of hypotheses to the feature level and profiling the two domains via these features can be a powerful new paradigm for improved understanding of toxicity from bioassay screening. In this talk, possible data mining methods to correlate screening data such as cell-viability assays with toxicity potentials will be presented.