| Mycobacterium can cause many serious diseases, such as tuberculosis and leprosy.Mycobacterium has a complex cell envelope and its virulence factors are distinct fromother pathogens. Especially, its membrane proteins play a critical role formultidrug-resistance and its tenacious survival ability. Knowing the subcellularlocalization of proteins and the types of membrane proteins will provide novel insightsinto understanding their functions and facilitate drug target discovery. In this study, anovel method was developed for predicting mycobacterial protein subcellularlocalization, membrane protein and their types by using over-represented tripeptides. Atotal of167mycobacterial proteins with distinct subcellular location,295and274non-membrane or membrane proteins were collected to evaluate the performance ofproposed method. Tripeptides were chosen for represent mycobacterial proteins. Thebinominal distribution was used to find the best feature set which can achieve thehighest accuracy. The support vector machine was applied for type’s prediction andmodel’s construction. The results of jackknife test showed that our method achieves anoverall accuracy of94.0%with an average accuracy of90.1%in distinguishingmycobacterial protein subcellular localization, an overall accuracy of93.0%with anaverage accuracy of92.9%in discriminating between mycobacterial membrane proteinsvs. mycobacterial non-membrane proteins, and an overall accuracy of93.1%with anaverage accuracy of83.6%in classifying mycobacterial membrane protein types. Bycomparing with other methods including different features, different algorithms andpublished methods, the proposed method showed excellent predictive performance.Based on the proposed method, a predictor called MycoMemSVM was built and can befreely available at http://lin.uestc.edu.cn/server/MycoMemSVM. It is anticipated thatMycoMemSVM will become a useful tool for the annotation of mycobacterialmembrane proteins and the development of anti-mycobacterium drug design. |
PDF Full Text Request