Aquatic Toxicity Prediction for Polar and Nonpolar Narcotic Pollutants with Support Vector Machines
Ovidiu Ivanciuc, BioChem Press, Ovidiu Ivanciuc · 2003
Motivation. Narcotic pollutants, that act by nonspecifically disrupting the functioning of cell membranes, are categorized as polar and nonpolar compounds. The toxicity prediction of narcotic pollutants with QSAR (quantitative structure-activity relationships) depends on the reliable determination of the mechanism of toxic action. The classification of the chemical compounds as polar and nonpolar narcotic pollutants based on structural characteristics is of utmost importance in predicting the aquatic toxicity for new chemicals. Method. Support vector machine (SVM) is a new machine learning algorithm that proved to be reliable in the classification of organic and bioorganic compounds. In this study we have investigated the application of SVM for the classification of 190 narcotic pollutants (76 polar and 114 nonpolar). Using an efficient descriptor selection algorithm, the energy of the highest occupied molecular orbital, the energy of the lowest unoccupied molecular orbital, and the most negative partial charge on any non–hydrogen atom in the molecule, all computed with the AM1 method, were found to be necessary for the discrimination of the polar and nonpolar compounds. The prediction power of each SVM model was evaluated with a leave–20%–out cross–validation procedure. Results. The classification performances of SVM models generated with the dot, polynomial, radial basis function, neural, and anova kernels, show that the statistical performances of SVM depend strongly on the kernel type and various parameters that control the kernel shape. An SVM model obtained with the anova kernel offered the best results, with three errors in calibration and four errors in prediction, all for nonpolar chemicals. Conclusions. SVM is a powerful and flexible classification algorithm, with many potential applications in molecular design, optimization of chemical libraries, and QSAR. In the present study we have demonstrated such an application for the identification of the aquatic toxicity mechanism.