Gene Regulation ▶ D3E



About D3E

Discrete Distributional Differential Expression (D3E) is a tool for identifying differentially-expressed genes, based on single-cell RNA-seq data. The main assumption we use in our method is that expression of RNA follows Poisson-beta distribution with three parameters α, β and γ. The first two correspond to the rate of RNA-polymerase binding and unbinding from a promoter region of a gene, and the latter corresponds to the rate of transcription when RNA-polymerase is in the bound state.

File format

Our software accepts a tab-separated read-count table, where rows correspond to genes, and columns correspond to individual cells. The file should have a header row which has the following tab-separated format:

"GeneID Label1 Label2 Label3 ... "

where Li are the cell type labels. Differential expression analysis can be performed on two cell types at a time. Please specify cell type labels of interest into the corresponding text boxes.

Each line should start with a gene ID, followed by a sequence of read-counts. Empty lines are ignored.


If you want your data to be normalised before analysis, please select 'Normalise' checkbox. The software performs normalisation method implemented in DESeq. If your data has been already normalised by any method please leave 'Normalise' checkbox blank.


D3E supports normalisation using spike-ins. To enable this feature please select 'Use Spike-ins' checkbox and make sure that the rows in the tabe, that correspond to RNA Spike-ins have labels starting with 'spike' (case insensitve).


The output table has the following columns:

  • Gene id : Id of a gene, that matches id in the input file.
  • α, β, γ : Parameter values of a fitted model.
  • s, f, d : Average burst size, expression frequency and duty cycle of a gene.
  • Rs, Rf, Rd : Log2 fold-change of the corresponding parameters.
  • p : p-value for a null-hypothesis, that two genes are not differentially expressed.
  • μ, Cv : mean and coefficient of variation of a gene expression in a corresponding cell type.

All the errors are listed below the results table.

Please be patient after the submission, as it may take a while to process the data. No progress information is given. In our experience it takes 5-10 minutes to process 20k x 100 matrix. For assistance please contact Martin Hember (