Application of multilayer perceptron for prediction of the rat acute toxicity of insecticides

With the growing number of insecticides that can potentially contaminate the environment, the determination of their acute mammalian toxicity is of prime importance in risk assessment. Chemoinformatics presents an alternative to animal testing because laboratory tests are costly in time and money and actively opposed by animal rights activists. In this work, the Quantitative Structure-Toxicity Relationship (QSTR) model established by using the artificial neural network (ANN) has been used for estimating the acute oral toxicity (LD50) of these insecticides to male rats. The 123 insecticides of the training set and the sixteen insecticides of external testing set have been described by means of using molecular descriptors. The QSTR model was validated internally and externally. A good results (Q2 =0.96 and Q2ext =0.95) were obtained. The prediction results are in good agreement with the experimental values of LD50.