Wondering what can cause skewing in an MAplot as seen below?

https://www.dropbox.com/s/wpittr9xsw4mjn7/MAplot1.png?dl=0

The MA plot for ChIP-seq is always going to look different than RNA-seq, for various reasons. Here on the horizontal you have comparable peaks, and the diagonal pointing up and right shows those peaks which are only present in whatever group is in the numerator of the LFC, with little to no signal in the group that's in the denominator. (For normalization for ChIP-seq, I could see making small alterations to the default RNA-seq approach to estimating size factors. But I don't have any code for this.)

As Michael Love says, the diagonals represent sites that are at or near zero on one condition but have higher binding in the other. What is unusual is that the most dense part of the plot lies on one of these diagonals (sites with binding in group1 but no binding in group2). The usual dense area in the middle of the plot, while less prominent here, is actually skewed a bit towards greater binding in group2. To see these two observations together is unusual.

It would be interesting to compare this to a non-normalized plot (set bNormalized=FALSE in the call to dba.plotMA). This would be to make sure that the normalization is not altering the binding characteristics too much. In RNA-seq, it is rare to see a large population of genes that are completely unexpressed in one group but expressed in the other, so normalization methods derived for RNA-seq can sometimes distort a ChIP-seq signal in ways that lead to misleading conclusions.

-R