Protein Subnuclear Localization Using a Hybrid Classifier Combined with Chou's Pseudo Amino Acid Composition

Protein subnuclear localization is very important for biology research and medicine design. Feature extraction and the classifier selection are two most important aspects in subnuclear localization prediction, the efficiency of their cooperation role directly decides prediction accuracy. Giving full play to the advantages of support vector machine in forecasting protein subnuclear localizations, in this study, jointly using multi-feature extraction of proteins and dimension reduction of laplacian eigenmaps, we designed a promising classifier system to predict protein subnuclear localization. Testing on the benchmark dataset which contained 714 proteins localized in nine subnuclear compartments, and applying jackknife test, our classifier system could obtain total accuracy of 81.23%, which was obviously higher than the results that had been obtained before, and the Matthews correlation coefficients of nine compartments were also higher than those of previous methods. All these success rates showed that our method was robust and effective for protein subnuclear localization, and it would be a supplement method for other protein classification and would play an important role in the fields of drug design and pharmaceutical industry.