[GenABEL-dev] ProbABEL: mmscore and interaction

[Yury Aulchenko]

What I remember for sure is that pre-adjusting for the interaction covariate and using "interaction-only" leads to deflated stats. So, this is not right way. 

If I remember correctly the best way is to include the interaction covariate in polygenic and then also keeping it in the design matrix when running mmscore. 

Yurii

Yurii Aulchenko

"L.C. Karssen" <lennart at karssen.org> wrote:
Hi Maksim,

 

 

 

On 18-10-14 03:13, Maksim Struchalin wrote:

> Hi Lennart,

>

> If ProbABEL is using the same 'two steps score test' as it was in the

> first version (section 'Two-step score test for association' in ProbABEL

> paper http://www.biomedcentral.com/1471-2105/11/134)

 

Yes it does. Nothing was changed there.

 

>  then, I guess, the

> GxE term can be included in the model and run in mmscore. It only should

> be assured that the interaction term goes to the second step (run the

> model trait_residuals ~ snp + snp*E in mmscore). Otherwise, it may

> results in false positives.

>

> Actually, in the first paragraph of section 'Two-step score test for

> association' of ProbABEL paper, it is written that "design matrix may

> include not only the main SNP effect, but e.g. SNP by environment

> interaction terms". So, I would include it and look what will happen.

>

 

Thanks! I'll have a look.

 

> P.S. IMHO it is interesting to run this scenario briefly in a small

> simulation study.

 

 

Good suggestion. I hope to have some time for that.

 

 

Best,

 

Lennart.

 

>

> Best,

> Maksim

>

>

>

> On 17/10/2014 21:54, L.C. Karssen wrote:

>> Dear list,

>>

>> There is a question on the forum [1] that I'm not sure how to answer.

>> The user wants to run mmscore and test for interactions. There is

>> nothing in ProbABEL that prevents her from doing that. However, in the

>> manual we state:

>>

>> "Though technically --mmscore allows for inclusion of multiple

>> covariates, these should be kept to minimum as this is a score test. We

>> suggest that any covariates explaining an essential proportion of

>> variance should be fit as part of GenABEL’s polygenic procedure."

>>

>> Does this also apply for interactions between SNP and a covariate? If so

>> we should produce an error message when both options are specified.

>>

>>

>> What is the proper way to deal with interactions and mmscore?

>>

>>

>> Best,

>>

>> Lennart.

>>

>>

>> [1] http://forum.genabel.org/viewtopic.php?f=10&t=896

>>

>>

>> _______________________________________________

>> genabel-devel mailing list

>> genabel-devel at lists.r-forge.r-project.org

>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel

>

>

>

> _______________________________________________

> genabel-devel mailing list

> genabel-devel at lists.r-forge.r-project.org

> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel

>

 

--

*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*

L.C. Karssen

Utrecht

The Netherlands

 

lennart at karssen.org

http://blog.karssen.org

GPG key ID: A88F554A

-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20141019/02f3bd15/attachment.html>