Walt Test 2 factors
1
0
Entering edit mode
julie • 0
@julie-23225
Last seen 4.1 years ago

Hi, I read many time DESeq2 doc(https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#differential-expression-analysis) specifically section 'Interaction'. There is a lot of great explanations. I see the example using a design very similar to what I use :

design ~genotype + condition + genotype:condition.

I also searched and read (https://rstudio-pubs-static.s3.amazonaws.com/329027_593046fb6d7a427da6b2c538caf601e1.html) that is very intersting but still haven't found the answer to my question.

I read also this(https://www.statisticshowto.com/wp-content/uploads/2016/09/2101f12WaldWithR.pdf) which is very close to what I need, but in this case it shows Wald test for 1 factors with 2 levels. And I want to know how calculation is done with 2 factors...

I read DESeq2 code, in particular about 'makeWaldTest' (github: https://github.com/mikelove/DESeq2/blob/master/R/results.R; lines 1031 to 1048) which is totally useful to see how Wald test is performed when you have 1 factor, like just a condition with for example Tumor versus Normal. but with those 2 doc and all my other searches , I still haven't figure out the answer for one question I have:

What I am searching for is : How is it calculated, how the Wald test is performed when you have 2 factors with 2 levels each, with a design like we have upper here? Like you want your comparison to be for example: ( Genotype1(Tumor/Normal) ) / (Genotype2(Tumor/Normal))

Many thanks for your help!

WaldTest_2factors DESeq2 • 1.4k views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 1 day ago
United States

The interaction term has a coefficient in the beta vector. So we test that estimated coefficient using the same formula as if the coefficient representing an LFC between two groups. The linear model framework makes the formula for testing these coefficient identical, although they conceptually represent different comparisons of group effects.

ADD COMMENT
0
Entering edit mode

Hi, thank you very much for this interesting answer. I did more reading and thinking after I read your answer to deeply understand. Would it be correct to say that it does a 'Multiple Linear Regression' to get the coefficients that you explain here, that are use for the following steps in the DESeq2? What I mean is if I understand correctly, for 1 factor with 2 levels, it does a linear regression to get the coefficient. So I would like to confirm if I understand correctly what is done when we use 2 factors with interaction, is it Multiple Linear Regression that is used to estimate coefficients? Thank a lot for your time and help again.

ADD REPLY
0
Entering edit mode

Hmm, well DESeq2 is a GLM but is conceptually similar to multiple linear regression. See the DESeq2 paper if you want more details on the model and how the parameters are fit.

ADD REPLY
0
Entering edit mode

Absolutely, I saw GLM in your paper. I was sure it applies to 1 factor. I understand is also for more factors also. Many thanks.

ADD REPLY

Login before adding your answer.

Traffic: 967 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