Question

DiffBind edgeR normalization --- why are the normalization factors all equal to 1?

0

Entering edit mode

tifaGis • 0

@00200f63

Last seen 12 days ago

United States

Hi,

I hope to try a full library size normalization but with edgeR.

So I performed the following:


data(tamoxifen_analysis)
tamoxifen = dba.normalize(tamoxifen, method = DBA_EDGER)
dba.normalize(tamoxifen, method = DBA_EDGER, bRetrieve = T)

Interestingly the normalization factors are all 1

$norm.method
[1] "lib"

$norm.factors
 [1] 1 1 1 1 1 1 1 1 1 1 1

$lib.method
[1] "full"

$lib.sizes
 [1]  652697  663370  346429  368052  466273  399879 1475415  616630  593224  706836 2575408

But apparently some normalization has happened as the ma plot is different from the one using raw data:

tamoxifen = dba.analyze(tamoxifen, method = DBA_EDGER)
dba.plotMA(tamoxifen, method = DBA_EDGER)

enter image description here

Could you please explain why is that? Thanks!

edgeR DiffBind • 1.1k views

ADD COMMENT • link updated 27 days ago by Harry • 0 • written 4 weeks ago by tifaGis • 0

0

Entering edit mode

It's interesting to see how the normalization factors are all 1, yet the MA plot reflects changes, indicating that the normalization process is still influencing the data! @retro bowl college

ADD REPLY • link 27 days ago Harry • 0

score 0 · Answer 1 · 2025-02-28

0

Entering edit mode

ATpoint ★ 4.8k

@atpoint-13662

Last seen 1 day ago

Germany

You perform library normalization which is expected to return normalization factors being all 1. If you want to use the more elaborate edgeR normalization then you have to tell that to the function, see ?dba.normalize. Looking at the MA-plot and the discrepancy between the loess (red) and neutral line (blue) indeed suggests that normalization is not optimal here. You're probably overestimating differences that are higher in the responsive group.

ADD COMMENT • link 4 weeks ago ATpoint ★ 4.8k

0

Entering edit mode

Hi, thanks for your response.

I'd like to know why

You perform library normalization which is expected to return normalization factors being all 1.

Shouldn't they be proportional to the actual library sizes?

I didn't want to do the more elaborate edgeR normalization, I just want to try the option of full library size + egdeR. In the tutorial, the author showed full library size + Deseq2 as the first and simplest normalization, where the norm factors are as expected (proportional to library sizes).


data(tamoxifen_analysis)
tamoxifen = dba.normalize(tamoxifen, method = DBA_DESEQ2)
dba.normalize(tamoxifen, method = DBA_DESEQ2, bRetrieve = T)


$norm.method
[1] "lib"

$norm.factors
 [1] 0.8099610 0.8232056 0.4298993 0.4567323 0.5786191 0.4962278 1.8309087 0.7652039 0.7361583
[10] 0.8771445 3.1959394

$lib.method
[1] "full"

$lib.sizes
 [1]  652697  663370  346429  368052  466273  399879 1475415  616630  593224  706836 2575408

ADD REPLY • link 4 weeks ago tifaGis • 0

0

Entering edit mode

No, normalization factors from edgeR rather than size factors from DESeq2 are not proportional, it's a different concept and implementation of the same end goal. The edgeR normalization factors are as expected.

ADD REPLY • link 4 weeks ago ATpoint ★ 4.8k