Prediction of mammalian toxicity of organophosphorus pesticides from QSTR modeling

Quantitative structure-toxicity relationship (QSTR) models were derived for estimating the acute oral toxicity of organophosphorus pesticides to male and female rats. The 51 chemicals of the training set and the nine compounds of the external testing set were described by means of autocorrelation vectors encoding lipophilicity, molar refractivity, H-bonding acceptor ability (HBA) and H-bonding donor ability (HBD) of the molecules. A feature selection was employed for selecting the most relevant autocorrelation descriptors. A PLS regression analysis and an artificial neural network (ANN) were used for deriving models accounting for the sex of the organisms in the estimation of the toxicity of pesticides. The best results were obtained with an 8/4/1 ANN model trained with the back-propagation and conjugate gradient descent algorithms. The root mean square residual (RMSR) values for the training set and the external testing set equaled 0.29 and 0.26, respectively.