Normalized counts of DESeq2 affected by design formula?
1
0
Entering edit mode
@blastzoneheimerdinger-15116
Last seen 5.6 years ago

Hello, sir.

I have a question regarding normalized counts of DESeq2 (Not VST or rlog transformed).

I used counts(normalized=TRUE) to get normalized data after DESeq() as follows. 

norm_tbl <- counts(dds, normalized=TRUE)

My question is that:

1. Can this normalized count table be used for downstream analysis, which I plan to use normalized count as dependent variable in modelling? Do this normalization account for normalizing for sequence depth variance?

2. Is this normalization affected by design formula when I created the DESeq dataset? I tried several different formulas, but no change can be observed for normalized counts.

Thanks for helping in advance.

 

deseq2 • 2.6k views
ADD COMMENT
5
Entering edit mode
@mikelove
Last seen 19 hours ago
United States

We recommend to use the variance stabilized data for downstream modeling e.g. for clustering or machine learning, etc. You can do:

vsd <- vst(dds)

Normalization is not affected by the design. It simply divides out the size factor for each column. See the DESeq2 paper for how the size factor is estimated (it doesn't use the design).

ADD COMMENT
1
Entering edit mode

 

 

Thank you very much for the reply!

I want to do generalized linear modeling or mixed modeling for downstream analysis, for covariate adjusting, is it the case for using VST also? As I read that VST values should not be used to differential expression analysis. My intend is that I want deseq2 to normalize the data for sequence depth or library size variance, and want to adjust covariates myself using model.

Because if VST performed, they seem to take care of covariates I want to adjust in my model in their function already, as different formula return different VST values. This part I do not understand.

As you suggested in DESeq2 for survival analysis, is this OK in this case to apply normTransform, then do modeling?

Thank you about normalization, I will read it again!

ADD REPLY
1
Entering edit mode

You should use the original counts in a GLM, not transformed counts. Also, the transformations do not adjust for covariates.

ADD REPLY
0
Entering edit mode

Hello, can I ask a clarification here? When you say "normalization is not affected by the design"

Are you saying that counts(dds, normalized = TRUE) is not affected by the design?

Or that vsd <- vst(dds) is not affected by the design.

I have previously seen that rlog(dds, blind = FALSE) "exposes" the design formula to rlog (meaning that this would be affected by the design).

Sorry if this has/is answered elsewhere, some of the terminology confuses me.

ADD REPLY

Login before adding your answer.

Traffic: 626 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6