Predicting Membrane Protein Types with Dimensionality Reduction and Kernel Method

The membrane protein type is an important feature in characterizing the overall topological folding type of a protein or its domains therein. How to fast and efficiently annotate the type of an uncharacterized membrane protein is a challenge. Some discrete models, such as DC (dipeptide composition) have been proposed to represent a protein sequence in the field of predicting membrane protein types. However, a high dimensional disaster may be caused by using this representation method. In this paper, a linear dimensionality reduction algorithm LDA (linear discriminant analysis) and a nonlinear dimensionality reduction algorithm KLDA (kernel linear discriminant analysis) are introduced to solve this problem by extracting the indispensable features from the high-dimensional DC space, respectively. Based on the reduced low-dimensional features, K-NN (K-nearest neighbor) classifier is introduced to identify the types of membrane proteins. As a result, experiment results show that using the proposed method to cope with prediction of membrane proteins types is very effective. It also can be seen that the success rates obtained by LDA are higher than those by other dimensionality reduction method such as PCA.