Classification and Modeling of Dose-Response Data from Phytotoxicity Experiments

摘要

Although numerous experiments have been conducted which deal with phytotoxic-ity, there is no general structure-activity relationship (SAR) approach for evaluating chemicalsaccording to their phytotoxicity. In this work, a methodology is presented to quantify phytotox-icity with effects data from the U.S. Environmental Protection Agency (USEPA) PHYTOTOX da-tabase. In the first step EC50 values are computed from dose-response relationships. Then anexperimental classification scheme is developed to accomplish a consistent toxicity index. Thechemicals were clustered using the Jarvis-Patrick algorithm. The clustering was performed bycomparing structural similarity using the "fingerprint" approach where bit patterns are pro-duced from substructures and compared using the binary Tanimoto distance norm. This proce-dure yielded a number of relevant classes of phytotoxic chemicals such as clusters ofphenoxy-type, s-triazine-type, or sulfonyl urea-type chemicals. In the final step, a quantitativestructure-activity relationship (QSAR) model was developed with this classification scheme. Thephenoxy class was chosen exemplarily and a model applied which is based on octanol/waterpartition coefficients and shape indices. Statistical verification yields a reasonable confidencelevel of the QSAR model.