View Single Post
Old 01-04-2013, 08:38 AM   #5
sdriscoll
I like code
 
Location: San Diego, CA, USA

Join Date: Sep 2009
Posts: 438
Default

I'm pretty sure in one of their papers they point out the possibility that there may be differential isoform expression but not differential locus expression. It is possible that there could be isoform switching or activation in a locus with significantly disturbing the total read count level of a locus. I would expect, however, that a differentially expressed locus would also have differentially expressed isoforms in it. Since cuffdiff models the variances separately for the isoform test and the locus level test sometimes it might report a single isoform gene as significant in one test but not the other. That I think is an unfortunate side effect of the statistics used and the variance estimations. The p-values are just a guide. Every gene may actually have different tolerances to differential expression. For some genes a two fold change could be significant regardless of the statistical test. It's important to keep that in mind.
sdriscoll is offline   Reply With Quote