We are hoping to extract differential promoter usage information from
our RNAseq data and were thinking about adapting DEXseq - using the count blocks
(using gtf files to filter for first exons or for last exons) and then applying the
statistical framework to identify differential usage. But before
working this through we were wondering if someone else had done so-
or if there is a good reason that this is a bad idea…

Hi Joan, 

I agree with Simon.

Just adding some additional cents, perhaps a challenge will be by the identification of first/last exons. I have seen that for genes with many transcripts, sometimes the first/last exon of one transcript is a middle exon for other transcripts. 

For example: 

http://www.ensembl.org/Homo_sapiens/Gene/Splice?db=core;g=ENSG00000139618;r=13:32315474-32400266

This cases are actually quite common in ENSEMBL transcript annotations and is challenging to infer that these exons are differentially used only because of differences in promoter usage. Thus the list of exons that you get will probably need further filtering steps. Ideally, one would like protocols that give you directly the information about start or termination sites (like here http://genesdev.cshlp.org/content/27/21/2380.long ), but I agree that your approach might be a good approximation. 

Alejandro

Yes, this should work.

Somehow, I'm sure that you would not be the first one to try precisely this because I have had several discussions in the past with colleagues thinking about similar ways of using DEXSeq -- but I can't find a paper on this right now. Still, I'm quite confident that it is worth a try.