A QSAR study of avian oral toxicity using support vector machines and genetic algorithms

This paper presents a study conducted on 116 pesticides to investigate the underlying relationship between structural properties and avian oral toxicity. Data were collected from several sources and screened to ensure quality and consistency. Physico-chemical and structural descriptors were obtained for the pesticides using the OpenMolGRID system. The resulting dataset was mined using principal component analysis (PCA), genetic algorithms (GA) and other classification algorithms. The final model was obtained using a support vector machine (SVM) combined with genetic algorithms for feature selection. It has good predictive ability (the error rate of the training set = 0.021, the error rate of the validation set = 0.158). Analysis of the descriptors indicates the prominent role of the interaction of pesticides with macromolecules and/or proteins in the mechanism of action.