Cuffmerge related query

[ParthavJailwala]

I have a query regarding what samples should be merged using cuffmerge, when you have multiple phenotypes (each with replicates). Lets say my mouse dataset comprises of one set of RNA-seq samples resulting from mutation of gene X, while there is another set resulting from mutation of gene Y. So the samples are:

Samples:
GeneXmutation (3 reps)
GeneYmutation (3 reps)
B6ctrl (3 reps)
MixedCtrl (3 reps)

The cuffdiff comparisons of interest are:

Phenotype A: GeneXmutation (3 reps) vs B6ctrl (3 reps)......[GeneX mutation is on the B6 background]
Phenotype B: GeneYmutation (3 reps) vs MixedCtrl (3 reps)....[GeneY mutation is on the Mixed background]

I have the cufflinks assemblies for each of the 12 samples. My question is: should the cufflinks assemblies (transcripts.gtf) of all the samples above be merged one time using cuffmerge (all 12 samples as above) OR I should be running two cuffmerges: one for Phenotype A comparison and other for Phenotype B comparison? Does one approach over the another make a huge difference?

Thanks

[sdriscoll]

Considering that when cufflinks assembles isoforms from alignments the proportion of novel features found compared to annotated features is small id say you could merge all of them and perform downstream analysis with that single merged annotation. If you think about it that's not much different than when we do expression analysis using a GTF from UCSC or Ensemble. You might find it convenient later in your project to have the expressions and DE tests of all of these samples in the same terms (ie relative to the same gene annotation).

[ParthavJailwala]

Hi Shawn,

Thanks for your input. I agree with your suggestion about (cuff)merging all the samples (from different phenotypes) together as one big merge and then use that merged.gtf for various cuffdiff comparisons of interest.
As you alluded to, I have observed that when different cuffdiff runs are based on the same merged.gtf, all cuffdiff output files have a) the same number of transfrags (both known + novel) and b) The XLOCid and TCONSid assigned by cufflinks are identical across DE lists, making cross-list lookups easier, specially for novel regions.

Thanks