[DESeq] baseMean

[Azazel]

I would like to ask how exacly baseMean is calculated in DESeq.

According to the manual from 2012-07-24 (http://tinyurl.com/8krvbny) the meaning of the three baseMean columns is:

baseMean mean normalised counts, averaged over all samples from both conditions
baseMeanA mean normalised counts from condition A
baseMeanB mean normalised counts from condition B

I interpret this as:

baseMean = (baseMeanA + baseMeanB) / 2

However I have this result from a DE test which I did with DESeq (only first entry shown):

id	baseMean	baseMeanA	baseMeanB	foldChange	log2FoldChange	pval	padj
geneA	4.6870306180134	3.72258134619134	6.1337045257465	1.64770194532405	0.720455295377725	0.137300747770432	0.461541670327437

(3.72258134619134+6.1337045257465)/2 == 4.928143
4.928143 != 4.6870306180134

There is obviously a discrepancy. Maybe I am interpreting it wrong.

I'd be thankful if someone could explain how I can replicate the calculation of baseMean.

P.S.: I asked a similar question in May this year. Simon nicely answered me that:

Yes, it is the normalized count, i.e., simply the number of reads divided by the size factor (normalization constant) of the sample

However, I do not know how I can get the "size factor" from the results table. If at all possible, I want to recompute baseMean given only baseMeanA and baseMeanB (is this even possible?). Because this way I wouldn't need to determine the sizeFactor.

[Simon Anders]

1. Maybe you have different numbers of samples in your two conditions. Then, the base mean will not be the average of the two per-condition means.

2. You real problem seems to be that you do not know how to get the size factors. If so, why do you not ask about this? Simply use the accessor function 'sizeFactors' on your CountDataSet to get a vector of size factors.