Design of a Novel Protein Feature and Enzyme Function Classification

Abstract

One of the most important researches in bioinformatics and biomedicine is to predict and classify the function of unknown protein. Recently, several studies based on alternative representation of protein have proposed for protein classification and prediction. However, most of these previous studies used only the predicted or global features extracted from protein sequence to assign function of distantly related proteins. Here, we describe a method that can assign enzyme function using features extracted from only protein sequence irrespective of sequence alignment. In our method, we design novel features presenting subtle distinction of local regions in protein sequence. In experimental results, the accuracy of the classifications for one-class versus one-class sub-problems is found in the range of 66.02% to 90.78% by support vector machine (SVM). Moreover, the results demonstrate that most of our features are valuable for enzyme function classification and add support to the facilitation of making discriminative feature set for specific enzyme function by combining traditional and novel features.