A method for improving the accuracy of predicting protein localization

In this paper, a system based on the novel Maximum Variance Projection (MVP) is proposed to improve the performance of protein subcellular localization prediction. Firstly, the protein sequences are quantized into a high dimension space using a new representation approach Position-Specific Score Matrix (PSSM). However, the problems caused by such representation are computation complexity and complicated classifier design. To sort out this problem, a new dimension reduction algorithm, MVP, is introduced. It extracts the essential features from the high dimension feature space. Then, K-Nearest Neighbor (K-NN) classifier is employed to recognize the subcellular localization of proteins according to the new features after dimension reduction. A good experimental result is obtained based on the jackknife dataset.