BioMed Central, Genome Biology, 10(11), 2010
DOI: 10.1186/gb-2010-11-10-r106
Nature Precedings, 2010
Nature Precedings
Full text: Download
High-throughput DNA sequencing is a powerful and versatile new technology for ob-taining comprehensive and quantitative data about RNA expression (RNA-Seq), protein-DNA binding (ChIP-Seq), and genetic variations between individuals. It addresses es-sentially all of the use cases that microarrays were applied to in the past, but produces more detailed and more comprehensive results. One of the basic statistical tasks is inference (testing, regression) on discrete count values (e.g., representing the number of times a certain type of mRNA was sampled by the sequencing machine). Challenges are posed by a large dynamic range, heteroskedas-ticity and small numbers of replicates. Hence, model-based approaches are needed to achieve statistical power. I will present an error model that uses the negative binomial distribution, with vari-ance and mean linked by local regression, to model the null distribution of the count data. The method controls type-I error and provides good detection power. I will also discuss how to use the GLM framework to detect alternative transcript isoform usage. A free open-source R software package, DESeq, is available from the Bioconductor project.