View-based3D Protein Structure Retrieval And Classification

Three-dimensional structure of protein is more conservative than the amino acidsequence itself, and can reveal protein function more, so it has a great significance in thestudy of protein function and interaction. Using computational methods to retrieve andclassify three-dimensional structure of the protein is an effective method to study proteinfunction and interaction. There are many kinds of feature extraction algorithm of proteins'three-dimensional structure; the molecular surface-based approach can effectively capture theshape of the protein surface characteristics. Feature extraction algorithm for three-dimensionalmodel based on the view is one of the most accurate methods. Improved, we will apply it inour experiments: First of all, we put the three-dimensional model into the center position,scaling and rotation; then, we project three-dimensional model to two dimensions, select thefront view, bottom view and left view, use five kinds of two-dimensional image featureextraction algorithm to extract characteristics of the three views. The five kinds of imagefeature extraction operators are the Fourier descriptors, Zernike moment descriptor, waveletdescriptor, PHOG descriptor and geodesic distance descriptors. At last, we connect five kindsof two-dimensional image features together as the characteristics of proteins'three-dimensional structure. Then, we use the L1algorithm to compute similarity measure.In the classified section, we use a multi-core learning algorithm to improve the classificationaccuracy. In addition, we also designed a three-dimensional protein structure retrievalsystem of human-computer interaction. It facilitates the scholars in this area to explorethree-dimensional protein structure and function, has a positive significance.