REACTIVE SUBSTRUCTURES IN THE PREDICTION OF AQUATIC TOXICITY DATA

Baseline QSAR models relating acute toxicity to log P(ow) are analyzed for predicting fish LC50 data of various compounds. Mostly (> 75%), t he high dependency of aquatic toxicity on lipophilicity resulted in ag reement of calculated and experimental data. Deviations occur with res pect to compounds revealing excess toxicity. From fathead minnow LC50 data for 500 organic chemicals, the toxicity of 122 compounds was foun d to be severely underestimated on the basis of log P(ow) alone. Among these 'outliers', 14 reactive substructures were identified, which ac counted for >73% of the compounds with excess toxicity. The applicatio n of a baseline QSAR supplemented by indicators for the reactive subst ructures also allowed the prediction of toxicity data based on other e ndpoints, indicating similar modes of interaction in various aquatic t ests.