Using Texture Descriptor and Radon Transform to Characterize Protein Structure and Build Fast Fold Recognition

Abstract

One of the most important research aims is to understand the relationship between structure and function of protein. Inspired by this motivation, automatic classification of protein structure becomes one of major research approaches. However, how to extract compact and effective feature to characterize protein structure is still a challenge to it. In this paper, 3-D tertiary structure of protein fold is mapped into 2-D distance matrix which can be further regarded as gray level image. Firstly, gray level co-occurrence matrix (CoM) of distance matrix image (DMI) is calculated and four descriptors based on it are taken as the first group of features. Next, DMI is transformed into projection view by Radon transform. In succession, the projection magnitudes are analyzed by histogram of which four central moments are taken as the second group of features. After that, we depict the structural meanings of gray distribution, various angles and pixels distance of CoM respectively, and determine the angle band used in Radon transform by the second derivation of the variance of the projections along different orientations. Finally, the presented feature extraction is validated by classification of 27 types of folds, compared with several feature methods based on sequence or structure. The results show that the presented method achieves significant improvement than other methods in terms of both low feature dimension and high classification accuracy.