Classification of Ligase Function Based on Multi-parametric Feature Extracted from Protein Sequence

摘要

One of the important goals of bioinformatics is to classify and predict the functions of proteins that have no sequence homolog of known functions. The purpose of this paper is to classify protein function by using multi-parametric feature, without sequence similarity. Firstly, we propose a method for generating novel features that present various local information of protein sequence based on positively and negatively charged residues. Then, we introduce a process of making optimal feature subset through combination of traditional and novel features extracted from protein sequence. Finally, we classify ligase enzymes by support vector machine (SVM). In experiment, only 375 out of 483 features were selected by feature selection, and the classification accuracy for 4th sub-classes in Enzyme Commission (EC) number is 98.35%. Our results demonstrate that most of novel features are valuable for specific enzyme function classification.