Sharing Information Across Genes to Estimate Overdispersion in RNA-Seq Data

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Steven Peder Lund*+ and Dan Nettleton
• Companies: Iowa State University and Iowa State University
• Keywords: RNA-seq data ; Generalized Linear Models ; Overdispersion

Abstract

Next Generation Sequencing technology can be used to measure gene expression (mRNA) levels. The resulting RNA-seq datasets consist of integer counts typically ranging from 0 to several thousand, with many observations less than 5. The low-count integers in RNA-seq datasets suggest using generalized linear models to identify differentially expressed (DE) genes. However, resulting GLM fits often provide evidence of overdispersion in many genes. While many methods exist for estimating overdispersion for a single gene, there are often few degrees of freedom available for these estimates. Here we present a new method for borrowing information across genes to obtain improved estimates of overdispersion parameters. Using these improved overdispersion estimates produces tests that more accurately control type I error rates and improve detection of DE genes.