SVM based approaches for classifying protein tertiary structures

Abstract

The tertiary structure of a protein molecule is the main factor which can be used to determine its chemical properties as well as its function. The knowledge of the protein function is crucial in the development of new drugs, better crops and synthetic biochemicals. With the rapid development in technology, the number of determined protein structures increases every day, so retrieving structurally similar proteins using current algorithms takes too long. Therefore, improving the efficiency of the methods for protein structure retrieval and classification is an important research issue in bioinformatics community. In this paper, we present two SVM based protein classifiers. Our classifiers use the information about the conformation of protein structures in 3D space. Namely, our protein voxel and ray based protein descriptors are used for representing the protein structures. A part of the SCOP 1.73 database is used for evaluation of our classifiers. The results show that our approach achieves 98.7% classification accuracy by using the protein ray based descriptor, while it is much faster than other similar algorithms with comparable accuracy. We provide some experimental results.