Predicting Protein Subcellular Localizations for Gram-Negative Bacteria Using DP-PSSM and Support Vector Machines

Abstract

Invisible bacteria are found almost everywhere, and having a great impact on our everyday life. Particularly, many species of gram-negative bacteria are pathogenic and cause a wide variety of diseases in humans and animals. It is crucial in drug design to cure diseases brought by gram-negative bacteria. Unfortunately, a new drug discovery can be expensive and time-consuming even with the advance of biotechnology. Designing a highly effective and efficient computational system, especially for identifying protein subcellular localization for gram-negative bacteria, is an important research field.In this paper, we propose a new computational system which combines a well-known classifier, support vector machines (SVMs), a protein descriptor, DP-PSSM (Directional Property-PSSM), and an optimal tool for system tuning. In addition, an evolutionary computation based feature selection technique is applied to further improve the performance of our computational system. Our computational system, EF-SVM-PSL, had been tested through 10 fold cross validation on predicting subcellular localizations of three gram-negative bacteria protein datasets, PS1444, NR828, and EV243. Our EF-SVM-PSL has a relative simple architecture and performs competitively with the best alternative systems.