Identifying Genes Differentially Expressed in Both of Two Independent Experiments

• Activity Number: 369
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Megan Orr and Peng Liu and Dan Nettleton*+
• Companies: North Dakota State University and Iowa State University and Iowa State University
• Keywords: bivariate analysis ; false discovery rate ; multiple testing ; q-value ; RNA ; gene expression

Abstract

We discuss two related problems: (1) estimating the number of genes differentially expressed (DE) in both of two independent experiments, and (2) identifying specific genes DE in both of two independent experiments while controlling false discovery rate (FDR). Scientists often use a two-step approach to address these problems. In step one, data from the two experiments are separately analyzed to produce a list of genes declared to be DE in each experiment. Usually, each list is produced using a method that attempts to control FDR within each experiment at some desired level alpha. In step two, the genes common to both lists are counted and identified as the genes DE in both experiments. A problem with this approach is that the resulting estimates of the number of genes DE in both experiments can vary greatly with alpha, and the value of alpha that produces the best estimate for any given pair of experiments is difficult to predict. Furthermore, little research has been done to evaluate how well this method controls FDR or ranks individual genes for significance when the goal is to find genes DE in both experiments. We describe new methods that address both these problems.