Question

logfoldStandard Error and wald-statistics in DESeq2 results

1

Entering edit mode

deena ▴ 20

@deena-7415

Last seen 6.6 years ago

Germany

Hi,

I am new to statitics and RNAseq analysis, I am using DESeq2 for the anaylzing the RNAseq data. I would like to know what lfcSE(logfoldchangeStandard Error) and stat (wald ) columns tries to convey. For an example. how can I interpret the following:

log2FC:-1.4; lfcSE: 0.144; stat:-10.13

Kindly guide me.

deseq2 wald • 18k views

ADD COMMENT • link updated 2.4 years ago by Michael Love 41k • written 9.1 years ago by deena ▴ 20

score 0 · Answer 1 · 2015-03-02

0

Entering edit mode

Michael Love 41k

@mikelove

Last seen 21 hours ago

United States

hi,

For information about the columns, read over the vignette:

vignette("DESeq2")

and the results man page:

?results

These both describe the columns.

ADD COMMENT • link 9.1 years ago Michael Love 41k

0

Entering edit mode

Dear Michael,

Thanks for your quick response. By going throught he vignette, I undersatnd that smaller the lfcSe more significant the effect of fold change? Am I right? What does that a value in wald statistic columns conveys.

Kindly guide me

ADD REPLY • link 9.1 years ago deena ▴ 20

0

Entering edit mode

The Wald statistic is the LFC divided by its standard error. This Wald statistic is used to calculate p-values (it is compared to a standard normal distribution) . So it's the ratio of LFC and SE which determines significance.

ADD REPLY • link 9.1 years ago Michael Love 41k

0

Entering edit mode

Thank you so much. Now I understood clearly. Also is the lesser the lfcSE, more singificant the log2fold change?

If basemean is the mean of normalized rowcounts, how log2foldchange is calculated.?

ADD REPLY • link 9.1 years ago deena ▴ 20

0

Entering edit mode

hi Deena,

You should look over our paper, if you want details about the method:

http://genomebiology.com/2014/15/12/550

ADD REPLY • link 9.1 years ago Michael Love 41k

0

Entering edit mode

Hi Michael,

Sorry to bring up an old post. I was reading your paper and noticed it says that the Wald statistic is a Z-score. Is this also true for the LRT statistic? If not, is it possible to convert it to a Z-score using scale() or are there considerations I need to consider? I am asking because I plan to meta-analyze different datasets.

Thanks a lot!

ADD REPLY • link 4.2 years ago NeuroGeneticsBoy ▴ 10

0

Entering edit mode

Only the Wald statistics are asymptotically Normal, not the LRT statistic. No, you should not need to adjust the Wald statistics.

ADD REPLY • link 4.2 years ago Michael Love 41k

0

Entering edit mode

Thanks so much for the advice. Can I substitute 0 for NA for Wald test statistic? Appreciate the help!

ADD REPLY • link 4.2 years ago NeuroGeneticsBoy ▴ 10

0

Entering edit mode

You can do what you want with these, thats up to you. They are 0 vs 0 genes.

ADD REPLY • link 4.2 years ago Michael Love 41k

0

Entering edit mode

Is ranking based on the Wald stat variable reasonable for gene ranking in the GSEA? Or ranking based on the lfcShrink (shrinkage Log fold change) better than stat? Thank you.

ADD REPLY • link 2.5 years ago zhilongjia • 0

1

Entering edit mode

I think both are reasonable. I haven't evaluated these against each other in a benchmark for GSEA.

The main difference between these is that lfcShrink estimates the effect size, where as n grows, the estimate converges to the parameter of interest (assuming we do not have bias).

As sample size grows, the Wald statistic does not converge to any parameter.

ADD REPLY • link 2.4 years ago Michael Love 41k