View Single Post
Old 04-04-2014, 08:16 AM   #2
Devon Ryan
Location: Freiburg, Germany

Join Date: Jul 2011
Posts: 3,480

Summing the counts from DEXSeq may lead to spurious results since you'll be double counting reads overlapping multiple exons. If you found any differential exon usage with DEXSeq, then you're even more likely to get spurious results here. You can get counts pretty quickly with featureCounts(), so give that a try.

If you're interested in "gender differences in response to some knock-out" then I would think you'd be most interested in the sex:strain interaction (so "design = ~ sex*strain").

The intercept is the base level of the model, which appears to be WT females (presumably you releveled() the strain factor in your design, since normally they're set alphabetically, meaning that KO females would be the base level), unless I have that backward. The "male vs female" result would be interpreted as "differences due to gender after accounting for differences due to strain". The "KO vs. WT" results would have a similar interpretation.
dpryan is offline   Reply With Quote