Kernel Independent Component Analysis-Based Prediction on the Protein O-Glycosylation Sites Using Support Vectors Machine and Ensemble Classifiers

Abstract

O-glycosylation means that sugar transferred to the protein. It can adjust the function of protein. To improve the prediction accuracy of O-glycosylation sites in protein, we used a new method of combining kernel independent component analysis with support vectors machine (KICA + SVM). The samples for experiment are encoded by the sparse coding with window size w = 51, 48 kernel independent components (feature) are extracted by kernel independent component analysis (KICA), then the prediction (classification) is done in feature space by support vector machines (SVM). The results of experiment show that the performance of KICA + SVM is better than that of KPCA + SVM, ICA + SVM, and PCA + SVM. Furthermore, we investigated the same protein sequence under various window size (w = 5, 7, 9, 11, 21, 31, 41, 51), and used the sum role to combine all the pre-classifiers to improve the prediction performance. The results indicate that the performance of ensembles of KICA + SVM is superior to that of pre-classifier. The prediction accuracy is about 90 %.