ABSTRACT
The purpose of this work is to develop robust and interpretable quantitative structure "activity relationship [QSAR] odels for assessing the aquatic toxicity of phenols using a combined set of descriptors encompassing the logP and recently developed variables [Monconn-Z variables]. The used dataset consists of 250 chemicals with toxicity data to the ciliate Tetrahymena pyriformis. For each compound, a total of 197 physico-chemical descriptors including logP and Molconn-Z descriptors were calculated. Multiple linear regression [LR] and Partial least squares [PLS] were used to obtain QSARs and the predictive performance of the proposed models were verified using external statistical validations. The results of stepwise-MLR analysis reveal that the AlogP, MlogP and ClogP models were not successful for the prediction of aquatic toxicity for all the compounds. And by using the logP [AlogP and MlogP] with Molconn-Z descriptors, the obtained QSARs were not successful enough until removal of some outliers. Two optimal QSARs were built with R[2] of 0.71 and 0.70 for the training sets and the external validation Q[2] of 0.69 and 0.68 respectively. All these selected descriptors in the best models account for the hydrophobic [AlogP, MlogP] and other electrotopological properties like SHCsatu, Scarboxylicacid, SHBa, gmax and nelem. PLS produces a good model using all the calculated descriptors with R[2] of 0.78 and Q[2] of 0.64, and hydrophobic and electrotopological descriptors show importance for the prediction of phenolic toxicity