Skip to main content
Figure 1 | Genome Biology

Figure 1

From: DGEclust: differential expression analysis of clustered count data

Figure 1

Information sharing between genes and between sample classes. The statistical model in DGEclust internally models the counts for each gene i in each library j as random variables sampled from a negative binomial distribution with gene-specific parameters μ i and ϕ i and gene- and experimental condition- (or tissue-) specific log-fold-changes β il . Different genes within the same condition l may share the same log-fold-changes, which are randomly sampled from discrete, condition-specific random distributions (G 1 and G 2 in the figure). This imposes a clustering effect on genes in each experimental condition; genes in the same cluster have the same colour in the figure, while the probability of each cluster is proportional to the length of the vertical lines in distributions G 1 and G 2. The discreteness of G 1 and G 2 is because they are random samples themselves from a Dirichlet process with global base distribution G 0, which is also discrete. Since G 0 is shared among all experimental conditions, the clustering effect extends between them, i.e. a particular cluster may include genes from the same and/or different experimental conditions. Finally, G 0 is discrete, because it too is sampled from a Dirichlet process with base distribution H, like G 1 and G 2. If the expression profiles of a particular gene belong to two different clusters across two experimental conditions, then this gene is considered differentially expressed (see rows marked with stars in the figure).

Back to article page