[BioC] DESeq Fold-Change calculation

[Lana Schaffer]

Simon,
Yes, I was incorrect in saying that the FC calculation used unnormalized values.

I have been given 2 equivalent datasets (8 controls and 8 treated) and need to 
"Show 'concordant' or reproducible hits".   Using DESeq, dataset B gave only 
21  (set B) transcripts significant compared to 431 for dataset A.  Turns out
That the stdev was higher for set B.
	Standard Deviation	
Dataset	control	treat
A	       76	     130
B	       96	     164

The package edgeR is giving more usable results for the 2 datasets :
416 (A) and 145 (B) transcripts with 76 overlapping.  Perhaps the variance
Is controlled better in edgeR?

Lana



-----Original Message-----
From: Simon Anders [mailto:anders at embl.de] 
Sent: Monday, August 27, 2012 2:11 AM
To: Lana Schaffer; bioconductor at r-project.org
Subject: Re: [BioC] DESeq Fold-Change calculation

Hi Lana

On 2012-08-26 13:16, Lana Schaffer wrote:
> I haven't checked to see if there has already been a discussion about 
> the fold-change Calculation in  DESeq,  but I have found that the fold-change is calculated using the
> Raw count values and not the normalized count values.   I usually use the fold-change
> With a cutoff value, but this isn't valid if the unnormalized counts are used for
> The FC calculation.   Any thoughts about this?

Of course, fold change is calculated using the normalized counts -- unless you have uncovered a very strange bug. Could you demonstrate with an example what you mean?

Cheers
   Simon