Extraction of 2D-Patterns from Protein Species Clustering

Abstract

Protein structure comparison and alignment is a difficult problem in computational biology because of the complexity in prediction and structural characteristics of proteins. In this paper we proposed feature extraction from Haralick features of pattern based protein species and feature selection method. The feature extraction is reliable method for clustering of given database from 2D-patterns using existing digital image processing techniques. The most important step in protein species clustering is feature selection. This feature selection method is Optimal Maximum Difference Feature Selection (OMDFS). This OMDFS is reliable and also it helps to improve the clustering accuracy. GLCM (Grey Level Co-occurrence Matrix) features are extracted from the protein images based on patterns. The extracted features are selected based on proposed OMDFS method. Experiments have been conducted on datasets of protein EF-hand fold images. Several feature selection methods are available. Accuracy of the model depends on the relevant feature selection. The proposed method selects only essential features and eliminates the irrelevant features. The experiment results show that the feature extraction model with proposed feature selection method improved classification accuracy.