Differential expression with RNASeq: length and depth does matter.

Ana Conesa, Gene Expression Lab, Centro de Investigaciones Principe Felipe, Valencia, Spain aconesa@cipf.es

• Abstract
  The RNA-seq technology has revealed in the last few years amazing new
  data on isoform and allelic expression, novel splice junctions, 3' UTR
  regions, antisense regulation and intragenic expression. RNA-seq is
  also increasingly being applied to the quantification of gene
  expression, as the number of mapped reads to a given gene or
  transcript is an estimation of the level of expression of that
  feature. How to process RNA-seq data to refine gene models or
  accurately portrait the dynamics of gene expression is today an area
  of active bioinformatics research. Although at the dawn of the RNA-seq
  applications it was claimed that this technology would render
  unbiased, ready-to-analyze expression data, the reality has turned out
  to be completely different and several shortcomings of the technology
  has been reported, such as the transcript length dependence of
  expression levels, or the influence of reads distribution in the
  assessment of differential expression. An underlying factor in RNA-seq
  analysis is the amount of reads generated in a given experiment. The
  more it is sequenced, the more transcripts are identified, the higher
  is the value of the expression level and the greater the power to
  identify differential expression. This leads to the question of how
  many reads should be generated in a RNA-seq experiment to obtain
  robust results.
  In this presentation we discuss the effect that sequencing depth has
  on the detection of expressed genes, the distribution of reads in
  transcript types and the number of differential expression calls. We
  show how sequencing depth influences the transcript length dependency
  of RNAseq. We present a novel approach for the analysis of count data
  that does not rely on parametric assumptions and shows robust
  behaviour for these biases.