You are here:
DEVELOPMENT OF QUANTITATIVE STRUCTURE ACTIVITY RELATIONSHIPS (QSARS) TO PREDICT TOXICITY FOR A VARIETY OF HUMAN AND ECOLOGICAL ENDPOINTS
Citation:
YOUNG, D. M., T. M. MARTIN, P. F. HARTEN, R. VENKATAPATHY, AND S. DAS. DEVELOPMENT OF QUANTITATIVE STRUCTURE ACTIVITY RELATIONSHIPS (QSARS) TO PREDICT TOXICITY FOR A VARIETY OF HUMAN AND ECOLOGICAL ENDPOINTS. Presented at The International Science Forum on Computational Toxicology, Research Triangle Park, NC, May 21 - 23, 2007.
Impact/Purpose:
To inform the public.
Description:
A web accessible software tool is being developed to predict the toxicity of unknown chemicals for a wide variety of endpoints. The tool will enable a user to easily predict the toxicity of a query compound by simply entering its structure in a 2-dimensional (2-D) chemical sketcher, entering a Simplified Molecular Input Line Entry System (SMILES) string, or entering a CAS number. The toxicity prediction models in the tool will be developed using both hierarchical clustering and genetic algorithm techniques. A hierarchical clustering based approach (based on Ward's method) will be used to divide an experimental toxicity dataset (for a given end point) into a series of structurally similar clusters. The structural similarity will be defined in terms of more than 3,000 physicochemical descriptors which include 2-D descriptors (such as connectivity and E-state indices) and 3-dimensional (3-D) descriptors (such as surface area and partial charges). A genetic algorithm based technique will be used to generate statistically valid quantitative structure-activity relationships (QSAR) models for each cluster (using the pool of descriptors described above). The toxicity for a given query compound will be estimated using the models generated from the clusters whose chemicals are the most structurally similar to the query compound since, in general, QSAR models yield better predictions if the test chemical is similar to the compounds used to build the QSAR model. Predictive models for several acute fish (96 hour fathead minnow LC 50) toxicity data sets have already been developed. The results indicate that our methodology can meet or exceed the predictive ability of literature models for these data sets. Future work involves using the prediction methodology to correlate additional toxicity end points.