Predicting Protein Subcellular Localization using PsePSSM and Support Vector Machines

Abstract

The prediction of protein subcellular localization (PSL) has been an important field of research. Many prediction systems nowadays have been developed that support the need. Most of these systems however focus on the development of new methods to describe a protein, which in turn can increase the prediction performance. In this paper,we propose a novel prediction system, an evolutionary algorithm based support vector machine for the prediction of PSL (ESVM-PSL) which aims to increase the prediction performance by optimizing SVMs. We apply ESVM-PSL to a set of proteins with jackknife validation. The prediction accuracy of SVMs is effectively increased from 48.9% to 67.5%. Our proposed method is also competitive with the previous systems in terms of prediction accuracy.