Aimed at solving continuous optimum parameter problems effectively in added drug design, this paper develops a novel ant algorithm termed continuous gridded ant colony (CGAC), where the spy ants are utilized to search the latent optimum grid in the domain completely and effectively. In order to test the effect, the CGAC algorithm was success in finding the best values of C and y, when the support vector machine (SVM) was used to fit the nonlinear relationship between the numerical representation of the chemical structure and IC50. The genetic algorithm (GA) was also used to obtain the appropriate feature subset simultaneously, because feature subset selection influences the appropriate kernel parameters and vice versa. The obtained results illustrate that GA-CGAC-SVM models have satisfactory prediction accuracy. The best quantitative modeling results in thirteen-descriptors model based on GA-CGAC-SVMr with mean-square errors 0.397, a predicted correlation coefficient (R2) 0.842, and a cross-validated correlation coefficient (Q^2) 0.756. The best classification result was found using SVM: the percentage (%) of correct prediction based on 7-fold cross-validation was 90.6%. The results demonstrate that the proposed CGAC algorithm provides a new and effective method to find the optimum parameters when the SVM tool is used.
展开▼