Using Multi-scale Glide Zoom Window Feature Extraction Approach to Predict Protein Homo-oligomer Types
The concept of multi-scale glide zoom window was proposed and a novel approach of multi-scale glide zoom window feature extraction was used for predicting protein homo-oligomers. Based on the concept of multi-scale glide zoom window, we choose two scale glide zoom window: whole protein sequence glide zoom window and kin amino acid glide zoom window, and for every scale glide zoom window, three feature vectors of amino acids distance sum, amino acids mean distance and amino acids distribution, were extracted. A series of feature sets were constructed by combining these feature vectors with amino acids composition to form pseudo amino acid compositions (PseAAC). The support vector machine (SVM) was used as base classifier. The 75.37% total accuracy is arrived in jackknife test in the weighted factor conditions, which is 10.05% higher than that of conventional amino acid composition method in same condition. The results show that multi-scale glide zoom window method of extracting feature vectors from protein sequence is effective and feasible, and the feature vectors of multi-scale glide zoom window may contain more protein structure information.