An Effective Feature Fusion Method for Protein Subnuclear Localization

Protein sub-nuclear localization plays a significant role in life science research. Feature extraction is the key step in the processes of protein sub-nuclear localization, it should include useful and sufficient protein information as possible. However, the single feature extraction method often can't extract sufficient information. Thus, we propose a novel method by combining PseAAC with PSSM to solve this problem. First, PseAAC and PSSM are used to extract the information, respectively. Secondly, the KPCA is used to reduce the dimensionality of PseAAC and PSSM. Thirdly, the useful information of PseAAC and PSSM will be integrated by proposed method. At last, the integrated feature vectors are classified by Random Forest. The classification performance will be evaluated with four evaluation indexes by jackknife test. According to the experiments results, the proposed feature fusion method is effective for protein sub-nuclear localization.