QSAR Modeling of Peptide Biological Activity by Coupling Support Vector Machine with Particle Swarm Optimization Algorithm and Genetic Algorithm.
Xuan Zhou,Zhanchao Li,Zong Dai,Xiaoyong Zou
DOI: https://doi.org/10.1016/j.jmgm.2010.06.002
IF: 2.942
2010-01-01
Journal of Molecular Graphics and Modelling
Abstract:A novel method coupling particle swarm optimization algorithm (PSO) and genetic algorithm (GA) was proposed to optimize simultaneously the kernel parameters of support vector machine (SVM) and determine the optimized features subset. By coupling GA with PSO, the particles produced in each generation in PSO algorithm were processed by crossover and mutation of GA, and then the particles could keep diversity to escape from local optima and find the global optima quickly and accurately. In order to evaluate the proposed method, four peptide datasets were employed for the investigation of quantitative structure–activity relationship (QSAR). The structural and physicochemical features of peptides from amino acid sequences were used to represent peptides for QSAR. The correlation coefficients (R) of training set of the four datasets were 1.0000, 0.9508, 1.0000, 0.9995, the R of test set of the four datasets were 0.9922, 0.9687, 0.9022, 0.7404, respectively. The root-mean-square errors (RMSEs) of training set of the four datasets were 0.0000, 0.0986, 0.0000, 0.0203, the RMSEs of test set of the four datasets were 0.2522, 0.2782, 0.9625, 0.2928, respectively. A protein dataset, which consists of 277 proteins, was also employed to evaluate the current method for predicting protein structural class, and the good results of overall success rate were obtained. The results indicated that the proposed method might hold a high potential to become a useful tool in peptide QSAR and protein prediction research.