More diff expressed genes than transcripts with Cuffdiff

[drosoform]

Dear all,

I am using newest version of Tophat -> Cuffdiff. I have two biological conditions, and 3 replicates for both conditions.

My results show ~200 something differentially expressed genes, but only 50 or so differentially expressed transcripts.

How is it that there are more differentially expressed genes than transcripts? For a gene to be deemed differentially expressed, wouldn't at least one of its transcripts have to be differentially expressed too? So I would have expect more transcripts than genes to be listed as differentially expressed, especially since some genes with multiple isoforms that are diff. expressed may sort of cancel out each other when looking at overall gene expression.

I noticed in the recent protocol published by Trapnell et al, they also showed about half as many transcripts as genes to be differentially expressed, in their example data set.

http://www.nature.com/nprot/journal/v7/n3/abs/nprot.2012.016.html

So there must be an explanation for this, I just don't know what it is. Please, could anyone explain? Sorry if this is something basic - I am very new to RNAseq!

[sdriscoll]

i think it's due to their variance modeling. i've had single isoform genes not called significant in the genes output but significant in the isoforms output. there should be no difference there.

[sepulveda]

I have noticed this problem, does anyone have a better answer? It does not make sense.