Gene expression data analysis using pseudo standard deviation minimization feature fusion method for cancer diagnosis.

Abstract

Over the past years applications of fusion technique have been growing rapidly. However, very few applications of the technique to microarray data have been reported. In this paper, we propose a new fusion method based on pseudo standard deviation minimization (PSDM) for the feature selection of microarray. This new method provides a more accurate set of features. Therefore the classification can be performed and functional meaning from the features can also be revealed. The new method is actually obtained through a combination of two different feature selection methods (FSMs). It is shown that it can explore nonperfect correlation between gene expression profile and cancer classes or feature detection algorithms. To evaluate its effectiveness, it is tested on lymphoma and leukemia microarray expression datasets and then compared with the existing methods. Self-organizing map (SOM) is used for feature classification. It can be seen through the comparison that the classification accuracy of the new fusion method is at least 2% ~ 3% higher than others.