A model-based dimension reduction approach to classification of gene expression data

The monitoring of the expression profiles of thousands of genes have proved to be particularly promising for biological classification, particularly for cancer diagnosis. However, microarray data present major challenges due to the complex, multiclass nature and the overwhelming number of variables characterizing gene expression profiles. We introduce a methodology that combine dimension reduction method and classification based on finite mixture of Gaussian densities. Information on the dimension reduction subspace is based on the variation of components means for each class, which in turn are obtained by modeling the within class distribution of the predictors through finite mixtures of Gaussian densities. The proposed approach is applied to the leukemia data, a well known dataset in the microarray literature. We show that the combination of dimension reduction and model-based clustering is a powerful technique to find groups among gene expression data.