Closed lcolladotor closed 9 years ago
This is the plot from the BrainSpan data that prompted all of this:
I think I found the culprit.... https://github.com/lcolladotor/derfinder/compare/1c5bc14667fbb6ac23006c7d52474cc6cc3891f3...master#diff-462d1cb76e572011166db2e74da8588fL203
Basically, I changed the default maxRegionGap
from 0L
to 300L
. I should be able to fix the results once the cluster is back online.
Actually, on a second thought fixing the results is not easy. The reason being that the widths and areas from the null regions are going to be incorrect
Script that fixes things as best as possible: https://gist.github.com/bf85e2c7d5d1f8197707
That is, it fixes the regions and annotation. However, the p-values are calculated with the available information from the incorrect null regions. In other words, the p-values are approximated and truly fixing them can only be done by re-running derfinder
Using https://gist.github.com/c0151c239af76a3972a9 you can visualize the error. Output at https://gist.github.com/56ddebd76a706a931874
Incorrect version:
Correct version:
I'm seeing weird results in Snyder and Hippo when reproducing the results from last year.
Hippo
New:
Old:
Snyder
New:
Old:
Colors of the DERs are not supposed to match perfectly as old versions use q-value < 0.10 and new versions use FWER < 0.05