Testing for Differences Between Multiple Groups in High-Throughput Sequencing Data Using Bayesian Multiscale Models

• Activity Number: 303
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Heejung Shim*+ and Ester Pantaleo and Matthew Stephens
• Companies: The University of Chicago and The University of Chicago and The University of Chicago
• Keywords: High-throughput sequencing ; RNA-seq ; ChIP-seq ; Bayesian multiscale model ; paired-end RNA-seq ; inhomogeneous Poisson process

Abstract

High-throughput sequencing technologies are now routinely applied at a genome-wide scale to collect a variety of phenotypic data. Testing for differences in these data between multiple groups is frequently encountered in genomics applications (e.g., eQTL mapping using RNA-seq and detecting differences in transcription factor binding across tissues using DNase-seq or ChIP-seq). Most approaches perform a test for differences using summary statistics such as total number of reads mapped to a gene or window although the original data consists of the counts of reads mapped to each base along the genome. Here we present statistical methods for testing for differences in the original data, which leads to increased power by using full information from the data. Specifically, our methods consider the data as an inhomogeneous Poisson process and test for differences in underlying intensities using Bayesian multiscale models. Unlike our previous Wavelet regression-based approach, the proposed methods can provide the power to detect differences with very few samples. We illustrate the proposed methods on DNase-seq data from 70 HapMap Yoruba LCLs and paired-end RNA-seq data.