chromVAR getMarkerFeatures() interpretation

[wuv21]

Hi! Question regarding how to interpret the values from chromVAR with ArchR. I followed the ArchR guide and created a MotifMatrix from running addDeviationsMatrix(). Afterwards, I ran getMarkerFeatures() to find the differential motifs between two annotated clusters with useSeqnames = "z" and useMatrix = "MotifMatrix". After looking at the returned object and filtering for motifs with FDR < 0.05, I noticed that there were a few motifs with a FDR < 0.05 but a log2FC of NaN. I'm a bit confused where the NaN is coming from, since I don't think this is a division by 0 or log(0) issue since it would return Inf or -Inf. Happy to provide more details if needed.

Thanks for providing ArchR to the community - have really enjoyed using it!

Thanks!
Vincent

[rcorces]

raised previously - #78

log2FC values do not have meaning for chromVAR. I think this has already been removed from the output in the most up to date versions of ArchR. What version are you using?

[wuv21]

Thanks @rcorces for the reply and for linking the previous post. Sounds good, I will use the meandiff to look at direction of change. Using ArchR 1.0.1

[wuv21]

I also notice NaNs from running getMarkerFeatures() on the regular PeakMatrix (no chromVar) - would you recommend using the meandiff here as well?

[rcorces]

Thanks for bringing this up. The previous check that was implemented was not correct and the Log2FC assay was not properly being set to NULL. I've fixed this via 48dccf3 on release_1.0.2

[rcorces]

@wuv21 - that doesnt sound right. I'm happy to take a look further if you are able to recapitulate what you see on the tutorial dataset. You can try upgrading to release_1.0.2 in case you are bumping into something that has already been addressed on this development branch

[wuv21]

Looks like it's all good now after upgrading to release_1.0.2. Thanks so much!