Using kernel SVM for predicting membrane protein types by fusing PseAAC and DipC

Abstract

In order to predict the types of membrane protein accurately, this paper firstly proposed a fusion feature representation, which contains a more comprehensive information of the original protein sequence by fusing two single feature expressions, pseudo amino acid composition (PseAAC) and dipeptide composition (DipC). Then, we proposed an improved support vector machine (SVM) method by introducing the idea of kernel function to evaluate prediction performance of the new fusion representation. In addition, we have deeply studied the influence of three different kernel functions as well as their kernel parameters on the prediction of membrane protein types. — Through experimental verification, it shows that the proposed integration representation with our improved SVM has a good performance in the prediction of membrane protein types. The final overall prediction accuracy can reach up to 89.64% under the Jackknife test method.