Identifying the mechanism of toxic action of selected compounds by artificial neural networks

In this study, classifying and predicting the non-polar narcosis, polar narcosis and reactive toxicity mechanism for 150 selected organic compounds were investigated using Artificial Neural Networks (ANNs). The variables used were the logarithm of octanol-water partition coefficients (logKow) and 10 quantum chemical parameters including the descriptors of energy, charge, and volume, which calculated with Gaussian 98. The 150 selected organic compounds were divided into two sets: training set (135 compounds) and test set (15 compounds). Supervised learning with backpropagation (BP) arithmetic was used. The results showed that the training error of network was smaller than 10–13, and 100% correct classification was achieved for test set.