Quantitative Structure-Activity Relationship study for pesticides by means of classification techniques

Abstract

The aim of this work was the comparison between κ-Nearest Neighbors (κ-NN) and Counterpropagation Artificial Neural network (CP-ANN) classification methods for modeling the toxicity of a set of 192 organochlorinated, organophosphates, carbamates, and pyrethroid pesticides measured as effective concentration (EC50). The EC50 values were divided into three classes, i.e. low, intermediate, and high toxicity. The 4885 molecular descriptors were calculated using the Dragon software, and then were simultaneously analyzed through κ-NN classification analysis coupled with Genetic Algorithms - Variable Subset Selection (GA-VSS) technique. The models were properly validated through an external test set of compounds. The results clearly suggest that 3D-descriptors did not offer relevant information for modeling the classes. On the other hand, κ-NN showed better results than CP-ANN.

Keywords

• Pesticidas
• κ-NN
• CP-ANN
• GA-VSS
• Teoría QSAR