I have some questions about Diffbind analysis when there are multiple groups to be compared. For example, if we have group1, group2 and group3 and wish to compare group1 vs group2, group1 vs group3 and group2 vs group3.

Are all the comparisons fully independent of each other except for the choice of analyzed regions based on peak calls? The concentrations for the different samples and groups give different results in different contrasts, so it looks like the data gets normalized within each contrast. It can be helpful to be able to compare concentrations across contrasts using the same source data used for the statistical analysis, which is not easy with the independent normalization. Is it possible to do the analysis using data normalized across all groups? Also, do the different contrasts use variance information from groups that are not used?

Thanks,

Vince

Up to Version 3.0, you are correct that DiffBind uses only the data associated with the samples in the contrast for each analysis.

The upcoming release of DiffBind will support the ability to include an arbitrary design over all of the data, and evaluate complex contrasts against that design. So one could for example have a design such as:

design = "~Replicate + Tissue + Condition + Treatment"

and set any of the design matrix coefficients however you like for testing, e.g.

contrast = c("Treatment","Drug1","Control")
contrast = c("Treatment","Drug2","Control")