Protein subnuclear location based on KLDA with fused kernel and effective fusion representation

Discriminated dimensionality reduction algorithm and informative feature representation are equal importance for improving prediction accuracy of protein subnuclear. Based on this thought, this paper simultaneously proposed an effective fused kernel function and an integrated feature expression for predicting protein subnuclear location. To obtain their optimal fusion parameter respectively, the particle swarm optimization (PSO) algorithm was employed to search them during the fusing processes. To verify the feasibility of our proposed approach, a standard public dataset was adopted to carry out the numerical experiment with k-nearest neighbors (KNN) as the classifier. The last results of Jackknife test method can be as high as 94.6779% with our fused kernel and representation, which undoubtedly reveals that our proposed integration method is of efficiency in protein subnuclear localization to a large extent.