Identification of secretory proteins based on similarity of amino acid sequences

Functioning as an “address tag” that directs nascent proteins to their proper sub-cellular localizations, signal peptides have become a crucial tool in finding new drugs or reprogramming cells for gene therapy. In the past twenty years, many algorithms have been proposed for predicting signal peptides. In spite of pioneering algorithms based on “scaled window” or “benchmark window”, similarity as another method for identification signal peptides is studied in details. By defining the similarity between every two sequences, we take the entire sequence effect into consideration, which is the important problem in modeling the whole amino acid sequence, and avoid the effect of variable-length problem. The normalized similarity represented the similarity among the overall sample is coincide with meaning of neighbours. So K-NN is adopted for classification. Then secretory proteins are identified based on similarity, similarity feature and SSLDA based reduced similarity features. And the prediction result for benchmark dataset is promising. The proposed similarity can also be used for other protein sequence analysis.