Prediction of Protein Backbone Structure by Preference Classification with SVM

Chang-Biau Yang, Kai- Yu Chen, Kuo- Si Huang
Proceedings of the 9th CONTECSI International Conference on Information Systems and Technology Management CONTECSI   unpublished
Given the primary sequence of a protein and its α-carbon coordinates, the allatom protein backbone reconstruction problem (PBRP) is to reconstruct the 3D coordinates of all atoms, including N, C, and O atoms on the backbone. A variety of methods for solving PBRP have been proposed, such as Adcock's method, SABBAC, BBQ, and Chang's methods. In this paper, we involve BBQ (Backbone Building from Quadrilaterals) and Chang's method as our candidate prediction tools. Then, we apply a tool preference
more » ... lassification with support vector machine (SVM) to determine which tool is more suitable for solving PBRP. According to the preference classification result, a proper prediction tool, either BBQ or Chang's method, is used to construct the atom of the target protein. Thus, after combining the results of all atoms, the backbone structure of the target protein is reconstructed. The three data sets of our experiments were extracted from CASP7, CASP8, and CASP9, which consists of 29, 24, and 55 proteins, respectively. The data sets contain only the proteins composed of standard amino acids. We improve the average RMSDs of BBQ results from 0.3955 to 0.3835 in CASP7, from 0.4437 to 0.4313 in CASP8, and from 0.4133 to 0.3691 in CASP9. Nowadays, the 3D coordinates of a protein can be built by X-ray crystallographic or nuclear magnetic resonance (NMR), but plenty of time and cost are required. Therefore, two computational methods, ab initio and homology modeling, were proposed for predicting the 3D protein structure (Zhang, 2008) . The ab initio method considers the interactions between the amino acids and the liquors to determine the stable status of each molecule. The homology method builds the templates with known protein structures and then produces a model based on the alignments between the target protein and the templates. The proper templates can be determined by the similarities of their
doi:10.5748/9contecsi2012/ps-434 fatcat:geumu2wcizeafaibig5nf2opty