Normalization: DESeq vs. EdgeR with method="RLE"

[aquila]

Hello all,

I am trying out different normalization methods for RNA-seq.
The newest release of EdgeR offers (among others) a normalization method called "RLE" that is supposed to be an implementation of what is also implemented in DESeq. This is explained in the edgeR manual. See ?calcNormFactors.

I thus expected to obtain the same normalization factors with both packages.
I am using part of the MAQC-2 data set that is referenced in a number of papers on RNA-seq. There are 14 samples, of which 7 are brain and 7 are UHR.

Here's what I tried:


"countsMatrix" is a matrix of raw counts (one column for each sample)
conds <- c( rep("brain", 7), rep("UHR", 7))

a)
cds <- newCountDataSet( countsMatrix, conds )
cds <- estimateSizeFactors( cds )
sizeFactors(cds)

b)
d <- DGEList(counts=countsMatrix, group=conds, lib.size=colSums(countsTable))
d <- calcNormFactors(d, method="RLE")
d$samples$norm.factors


Results:
a)
1.1430 1.1597 1.1695 1.1707 1.1751 0.3293 1.1643 1.1489 1.1650 1.1781 1.1802 0.4877 1.1617 1.1596

b)
1.0546 1.0354 1.0167 1.0291 1.0178 0.7133 1.0330 1.0645 1.0705 1.0876 1.0837 0.7619 1.0796 1.0564


Might anyone have a suggestion why the resulting normalization factors are different?

[Jouneau Luc]

Hello aquila,

would you please tell me where did you download this MAQC dataset ?

Thanks in advance

Luc

[aquila]

The data is available on

http://www.ncbi.nlm.nih.gov/sra

Search for: SRA010153

There are several data sets there. I downloaded the following:

SRX016368 (7 samples)
SRX016366 (7 samples)


I found these data sets referenced in the following publication (among others):
Bullard et al, Evaluation of statistical methods for normalization and differential expression in mRNA-Seq experiments, BMC Bioinformatics 2010, 11:94.

[ning]

Hi aquila

http://answerpot.com/showthread.php?...zation+factors

https://stat.ethz.ch/pipermail/bioc-sig-sequencing/2011-February/001810.html

For edgeR, I guess you need to multiply the output of calcNormFactors() by the library size to get the "normalization factor"?