Hi,

Is there an easy way to visualize the csaw results of differential binding as MA plot or volcano plot ?

Thanks!

I will only add that all of the solutions presented here are operating on per-window statistics. This is reasonable enough but it is often desirable to generate plots on per-region statistics, given that we interpret the results in terms of genomic regions rather than windows.

There are several ways to do this depending on how you want to obtain a region-level summary statistic. Most workflows involving the current version of csaw (1.20.0) will use the window with the lowest p-value as the representative window to obtain region-level statistics. In the BioC-devel version of csaw (1.21.4), combineTests and friends will automatically report a representative window based on the statistical properties of each combining procedure. (Of course, you can continue to use the window with the lowest p-value as the representative if you so choose.)

The most extreme case would be to repeat the counting with regionCounts() over the interval spanned by each region, repeat the DE analysis and obtain genuine region-level statistics. I don't do this much because interesting log-fold changes in DB subintervals of a region can be "diluted" towards zero by including counts for the non-DB remainder of the region; but mostly because it's a hassle to have to do all that again.

I have a function for MA-plots that will use smoothScatter() to visualize the results while coloring regions below a p-value cutoff in a separate color. Say you have your results (typically the output of topTags) called tt which contains FDR, logCPM and logFC, then you can do:

smoothMAplot(X.value = tt$logCPM, Y.value = tt$logFC, P.values = tt$FDR, Sig.Thresh = 0.05)

This will give you a plot like this. The legend indicates how many genes are below the FDR cutoff per direction and what the minimal fold change is that is still significant.

csaw uses edgeR to do the differential testing, so you can use plotMD() to make MA plots (aka mean-difference plots) just like in edgeR.