cummeRbund get significantly differentially expressed genes

[Olioli]

Hello,

I am conducting an RNA-Seq analysis to identify significantly differentially expressed genes (SDEGs) between 5 conditions. I followed the Tuxedo protocol which is tophat/cufflinks/cuffmerge/cuffdiff/cummeRbund (http://www.nature.com/nprot/journal/...t.2012.016.pdf) and I am at the final part where I need to determine the number of SDEGs.
I followed the cummeRbund instructions in the manual (http://compbio.mit.edu/cummeRbund/manual_2_0.html) and here's my output:

>cuff_data<-readCufflinks(genome="genome.fa",gtfFile="merged.gtf")
CuffSet instance with:
	 5 samples
	 16011 genes
	 36460 isoforms
	 24184 TSS
	 15312 CDS
	 160110 promoters
	 241840 splicing
	 135440 relCDS
>gene.diff<-diffData(genes(cuff_data))
> sig.gene.diff<-subset(gene.diff, significant=="yes")
> nrow(sig.gene.diff)
[1] 4500
> mySigGeneIds<-getSig(cuff_data,alpha=0.05,level='genes')
> length(mySigGeneIds)
[1] 1386
> mySigIsoformIds<-getSig(cuff_data,alpha=0.05,level='isoforms')
> length(mySigIsoformIds)
[1] 900

So when I use diffData() and filter only significant results using subset() I get a total of 4500 SDEGs (this is for all pairwise comparisons that I can extract separately using another subset() with sample names) BUT when I use the getSig() method with alpha=0.05, I only get 1386 gene IDs (not 4500), so that's less than a third from what I get with diffData. I tested at "isoform" level to see if it would match but it didn't.
The part in the cummeRbund manual about getSig() says "a alpha value can be provided on which to filter the resulting list (the default is 0.05 to match the default of cuffdiff)."
From what I understood, the diffData() method is supposed to just extract the information from cuffdiff without making additional tests so shouldn't I get the same number of genes?
I would be greatful if someone could explain me clearly the difference between those two methods and which one should be used to determine SDEGs.
Thanks a lot for your help,
Oli

[Olioli]

I realized that if I use the getSig() method one pair of conditions at a time with the command:
cond1_vs_cond2_getSig<-getSig(cuff_data,x="cond1",y="cond2",alpha=0.05,level="genes")
and sum the length of the vectors (=number of SDEGs) for the 10 possible pairwise comparisons (I have 5 conditions) I get a total of 4773 SDEGs.
So I think that, with 5 conditions, using getSig(cuff_data,alpha=0.05,level='genes') will correct for multiple testing going through all conditions and this is why I get a smaller number of genes, so I shouldn't use it to determine SDEGs for a particular pair of conditions, can someone comfirm this?
On the other hand I am still not sure whether I should use diffData() + filter only significant results + filter by pairwise comparisons OR if I should use getSig(cuff_data,x="cond1",y="cond2",alpha=0.05,level="genes") for each pair of conditions.
I thank anyone who may help me with this!!

[Olioli]

After more than 400 views nobody answered or helped me at all,
I am sure that someone should be able to help me on this forum, I will be glad to answer any question you may have.