Applications of penalized mixture distributions to microarray data analysis

Abstract

The main goal in analyzing microarray data is to determine the genes that are differentially expressed across two types of tissue samples or samples obtained under two experimental conditions. In this paper we propose a penalized normal mixture model (PMMM) to estimate the parameters within the framework of maximum likelihood We penalized both the variance and the mixing proportion. The variance was penalized so that the log-likelihood will be bounded, while the mixing proportion was penalized so that we can apply the modified likelihood ratio to test for the number of components. Additionally, a weight function was introduced because the estimation method is sensitive to the presence of statistical outliers. Simulation study is conducted to demonstrate effectiveness of PMMM. Finally, the penalized method is applied to the rat data for genes in middle ear mucosa of rats with and without subacute pneumococcal middle ear infection. © Dynamic Publishers, Inc.