Principal Change Analysis for Comparing Multiple ChIP-Seq Profiles in Two Biological Conditions

• Activity Number: 59
• Type: Topic Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Hongkai Ji*+
• Companies: The Johns Hopkins University
• Keywords: ChIP-seq ; next-generation sequencing ; principal component analysis ; multiple testing ; genomics ; gene regulation

Abstract

ChIP-seq is a powerful approach to create genome-wide profiles of protein-DNA interactions. Early applications of ChIP-seq mostly involve finding protein-DNA binding sites in a single biological condition. As applications of the technology diversify and more data become available, the data structure in ChIP-seq studies becomes increasingly more complex. This brings new statistical and computational challenges for data analysis. A new and common situation many investigators face is to compare two biological conditions (e.g., two different cell types or tissues), each has ChIP-seq profiles for multiple different proteins (e.g., different histone marks H3K4me3, H3K27me3, H3K36me3, etc.), and each protein has a few replicate samples to measure biological variability. The questions of interest are what genes are different between the two conditions, and how they are different. We develop a statistical framework to address this issue. Our approach first identifies principal difference patterns in the genome. It then tests whether each gene is different along the principal difference directions. We show statistical properties of this method and illustrate its applications in real data.