Identifying nuclear protein subcellular localization using feature dimension reduction method

The subcellular location of a protein is closely correlated to its function. Facing the deluge of protein sequences generated in the post-genomic age, it is necessary to develop useful machine learning tools to identify the protein subcellular localization. DR (Dimensional Reduction) method is one of most famous machine learning tools. Some researchers have begun to explore DR method for computer vision problems such as face recognition, few such attempts have been made for classification of high-dimensional protein data sets. In this paper, DR method is employed to reduce the size of the features space. Comparison between linear DR methods (PCA and LDA) and nonlinear DR methods (KPCA and KLDA) is performed to predict subcellular localization of nuclear proteins. Experimental results thus obtained are quite encouraging, which indicate that the DR method is used effectively to deal with this complicated problem of viral proteins subcellular localization prediction. The overall jackknife success rate with KLDA is the highest relative to the other DR methods.