Schematic illustration of two alternative strategies (gene-based and exon-based) for detecting differential expression between two sample groups.

<p>The RNA-seq data are from the MAQC dataset, containing two types of biological samples: human brain reference (brain) and human universal reference RNA (uhr). (<b>A</b>) Exon structure of the gene DCUN1D5 (<b>B</b>) Separate read counts for the eight exons of the gene. (<b>C</b>) Normalized total read counts across all the exons for the gene. (<b>D</b>) Logarithmic (base 2) fold change between the sample groups separately for each exon. The number of stars above a bar indicates whether one or both of the two software packages (limma, edgeR) identify the particular exon as significant at <i>p</i><0.05. (<b>E</b>) Gene-level log fold change between the sample groups obtained using directly the gene-level read counts (gene-based strategy; left bar) or by taking the median over the exon-level changes (exon-based strategy; right bar). The exon-based strategy supports differential expression (median <i>p</i> = 3.69e–06 and 1.64e–09 with limma and edgeR, respectively), whereas the conventional gene-based strategy suggests that the gene is equally expressed in both groups (<i>p</i> = 0.91 with both limma and edgeR). The fold changes were determined here using the limma software package.