Reducing the Variation of Gene Expression Patterns: A Grey Model Approach Applied to Microarray Data Classification

Abstract

The gene expression levels measured from microarray spots vary between patients with the same type of tumor. This reduces the performance of microarray classification, especially when the microarray dataset has few samples. Here, we introduce the grey model GM(1,1) of the grey system theory for modeling gene expression patterns of small samples to eliminate variations. To evaluate the application of GM(1,1), we have combined GM(1,1) with GA/MLHD approach to solve the problem of multi-class classification. The GM(1,1)- GA/MLHD model was tested on two published microarray datasets: (1) NCI60 cancer cell lines and (2) the GCM dataset. The experimental results show that the GM(1,1) gave gene expression patterns with less variations and helped the MLHD classifier to improve classification accuracy over the method of GA/MLHD, but they also outperformed many class prediction approaches.