Weird-looking plots about idr

[DarwinAwardWinner]

Here's a link to example files that priduce plots like the above: https://www.dropbox.com/sh/k2193eqe1j8qun9/AAASAJG9BkzXHXPDHKdlLVhha?dl=0

[DarwinAwardWinner]

Ok, I think I know what the problem is. The peak caller I'm using (MACS2) is returning lots of identical enrichment scores, which means that peaks with those scores are essentially sorted randomly, throwing off the IDR algorithm. The patterns of identical scores exactly match the stair-step patterns seen in the top plots.

[DarwinAwardWinner]

Here's a look at an example plot for one sample's peak call scores vs rank:

[DarwinAwardWinner]

It turns out that the answer was to use the -log10(p-value) column instead of score or signal value, since this column seems to have the greatest number of unique values for MACS2. In contrast, for Epic, the column with the most unique values is score. So the lesson is to look out your peak output and figure out which potential ranking column has the fewest duplicates.

[DarwinAwardWinner]

Also, I think the above plots look weird partially because all the red points have black outlines. So in areas of high point density, the red points look black because all you see are the black outlines.

(Also also: MACS2 outputs up to millions of peaks if you let it, so one should filter to only the best 150k or so, or else idr will take forever to run.)