Analysis of a Probe-Level Linear Mixed Model for Oligonucleotide Arrays

• Activity Number: 547
• Type: Contributed
• Date/Time: Thursday, August 10, 2006 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Alexander Cambon*+ and Caryn Thompson and Brian Wattenberg
• Companies: University of Louisville and University of Louisville and University of Louisville
• Keywords: probe ; mixed models ; microarrays ; oligonucleotide ; affymetrix ; gene expression

Abstract

As the number of replicates and treatments in microarray experiments increases, mixed models are becoming more widely applicable as a method for assessing statistical significance of gene expression data. In this paper, a randomized complete block design consisting of 21 hgu133plus2 microarrays, 7 treatments and three replicates is analyzed using the methods outlined in Chu et al.'s 2002 paper. This method includes probe as a fixed factor and array as a random effect. The results are compared with those obtained using quantile normalization, median polish, and two-way ANOVA. One advantage of the probe level model is that probe variation within probe set is accounted for by including probe as a factor. This increases the sensitivity of the test for treatment effect. The probe level model also increases the sample size for residual analysis and outlier detection of influential arrays.