Authors: Cheng-Wei Hsieh; Hui-Huang Hsu; Tun-Wen Pai
Addresses: Department of Computer Science and Information Engineering, National Central University, Jhongli City, Taoyuan County 32001, Taiwan (R.O.C.) ' Department of Computer Science and Information Engineering, Tamkang University, New Taipei City 25137, Taiwan (R.O.C.) ' Department of Computer Science and Engineering, National Taiwan Ocean University, Keelung 20224, Taiwan (R.O.C.)
Abstract: To determine the structure of a protein by X-ray crystallography, the protein needs to be purified and crystallized first. However, some proteins cannot be crystallized. This makes the average cost of protein structure determination much higher. Thus it is desired to predict the crystallizability of a protein by a computational method before starting the wet-lab procedure. Features from the primary structure of a target protein are collected first. With a proper set of features, protein crystallizability can be predicted with a high accuracy. In this research, 74 features from previous researches are re-examined by two filter-mode feature selection methods. The selected features are then used for crystallization prediction by three versions of AdaBoost. The Support Vector Machines (SVMs) are also tested for comparison. The best prediction accuracy of AdaBoost reaches 93 percent and 48 important features are identified from the collected 74 features.
Keywords: X-ray crystallography; protein crystallization; feature selection; SVM: support vector machines; AdaBoost; data mining; bioinformatics.
International Journal of Data Mining and Bioinformatics, 2013 Vol.7 No.2, pp.214 - 227
Available online: 29 Mar 2013 *Full-text access for editors Access for subscribers Purchase this article Comment on this article