Hi,

I used DiffBind to count the ChIP-seq reads (dba.count) and plot the correlation heat map (dba.plotHeatmap), and a have a question. By default dba.plotHeatmap uses all the samples and computes the correlations between every pair, and I would like to compute and plot the correlations only among the binding differences. For example my samples are A_u (untreated sample), A_t (treated sample), B_u, B_t, C_u, C_t, etc. I would like to compute first the binding differences A_t - A_u, B_t - B_u, C_t - C_u, etc. and then to use dba.plotHeatmap or something similar to plot the correlation heat map and see which samples behave similarly. Could you please tell me if this is possible using DiffBind?

Thank you,

Razvan

Hi Razvan-
   
The plotting functions dba.plotHeatmap() and dba.plotPCA() take a parameter, sites, that allows you to narrow down what peaks to use in the plot. You can get the DB sites using dba.report().

Here is an example ofthe steps using the supplied data object as an example. Suppose we want to look at how all the samples cluster using the DB sites obtained from an analysis that doesn't use the MCF7 sites:
   
data(tamoxifen_counts)
tamoxifen
# Contrast Resistant non-MCF7 with Responsive non-MCF7 samples
tamoxifen <- dba.contrast(tamoxifen,
                        group1=tamoxifen$masks$Resistant&!tamoxifen$masks$MCF7,
                        group2=tamoxifen$masks$Responsive&!tamoxifen$masks$MCF7)
tamoxifen <- dba.analyze(tamoxifen)
tamoxifen
# Retrieve DB sites in a GRanges report
tamRep <- dba.report(tamoxifen,contrast=1,th=.01)
# get DB sites from report
sitevec <- as.integer(names(tamRep))
# compare plots
plot(tamoxifen,main="All Sites")
plot(tamoxifen, sites=sitevec,main="DB Sites")
par(mfrow=c(1,2))
dba.plotPCA(tamoxifen,sub="All Sites")
dba.plotPCA(tamoxifen, sites=sitevec, sub="DB Sites")

I'm looking at making this a little more straighforward by making it so you can just pass in hte report without creating the sitevector.

-R