Nonparametric Analysis of Microarrays for Identifying Reproducible Differentially Expressed Genes

• Activity Number: 410
• Type: Contributed
• Date/Time: Thursday, August 15, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section*
• Author(s): Jeanne Kowalski*+ and Giovanni Parmigiani and Marianna Zahurak and Christopher Umbricht and Martha Zeiger
• Affiliation(s): Johns Hopkins University and Johns Hopkins University and Johns Hopkins University and Johns Hopkins University and Johns Hopkins University
• Keywords: Gene expression data analysis ; microarrays ; multiple testing

Abstract

A focus in genomic research is to relate expressions from thousands of genes to phenotype--for example, cancerous status. Replicated microarray data enables estimation of the variability in expression within a gene. With the high cost of microarray technology, the number of replicates is typically small, often two, raising issues on how to best to combine the limited information for identifying differentially expressed genes. We propose a conditional characterization approach in which we first identify a set of reproducible genes and then further partition this set into candidate differentially expressed genes. Depending upon the randomization scheme implemented, assumptions are made regarding the distributional properties of intensities. The connection between the proposed approach with existing descriptive methods, including SAM, is discussed. The method is applied to nylon filter data, where the problem of identifying genes as potential disgnostic tools in thyroid cancer is examined.