Prediction of aqueous toxicity for heterogeneous phenol derivatives by QSAR

We provide QSAR models for the growth inhibition of the ciliated protozoan Tetrahymena pyriformis by 250 mechanistically diverse phenolic compounds. The simultaneous linear regression analysis on 1338 topological, geometrical, and electronic molecular descriptors over 200 molecules leads to a seven-parameter relationship with R = 0.851 and leave more out Rl − 60% − o = 0.730, while a model based on flexible descriptors improves to R = 0.880 and Rl − 60% − o = 0.812. An external test set of 50 related derivatives demonstrates that both models show good predictive power with rms = 0.418 and rms = 0.352, respectively, comparing fairly well with previously reported Artificial Neural Networks with similar rms. Finally, we employ the best QSAR equation to estimate the unknown aqueous toxicity of 74 structures.