QSARs for toxicity of DDT-type analogs using neural network

Structure-toxicity relationships for 120 insecticidal DDT-type molecules including diaryl nitropropanes (Prolan analogs), diaryl trichloroethane and other DDT isosters, collected from different literature sources, to Musca domestica. were analysed by regression analysis (RA) and a neural network model (NN). The steric factors are extremely important to the toxicity of all the DDT-type analogs. The lipophilicity is also important for the groups, because it facilitates delivery of these neurotoxicants to the site of action in the nerve. On the basis of training results, the NNs proved to give better results than a regression analysis technique and the most accurate predictions. To describe the role of each of the descriptors we suggested a new method based upon the estimation of the connection weights, the identification and the rationalisation of the residuals.