Understanding probabel GxE interaction output

Questions about ProbABEL are welcome here.
Forum rules
Please remember not to post any sensitive data on this public forum.
The first few posts of newly registered users will be moderated in order to filter out any spammers.

When get a solution to the problem you posted, please change the topic name (e.g. from "how to ..." to "[SOLVED] how to ..."). This will make it easier for the community to follow the posts yet to be attended.
agiri_genepi
Posts: 2
Joined: Thu May 21, 2015 8:13 pm

Understanding probabel GxE interaction output

Postby agiri_genepi » Thu May 21, 2015 8:36 pm

Hello, I'm new to GenABEL and ProbABEL. I'm using ProbABEL at the moment, however, I'm a little confused about the output.

Say for example I'm fitting the following model:
y ~ m + cov1 + cov2 + E + SNP + SNP*E ------ (model 1)

What I'm really interested in is testing if the SNP*E is statistically significant. Normally, I would do a likelihood ratio test by subtracting the -2 log likelihoods of the full model (model 1) with a reduced model (model 2 below) to obtain a chi-square statistic with 1 degree of freedom.

y ~ m + cov1 + cov2 + E + SNP ------ (model 2)

I fit the full model (model 1). The output for ProbABEL provides a chi2_SNP statistic. I'm assuming that this is the chi-squared statistic from the LRT comparing the full model as I fit above with a null (no covariate) model (NOT to be confused with the reduced model)... is this correct? Verification of this would be helpful.

So if I wanted a p-value for the SNP*E interaction term comparing nested models (full - reduced) to get the LRT, how would I go about getting it with ProbABEL? (other than simply doing beta_SNP*E/se_SNP*E, which would be equivalent to doing a wald test)

Is it appropriate to fit model 1 then model 2 and then subtract the chi2_SNP values to get the SNP*E chi2 for 1 degree of freedom?

Thank you for your help.

wentingwu
Posts: 1
Joined: Fri May 29, 2015 10:35 pm

Re: Understanding probabel GxE interaction output

Postby wentingwu » Fri May 29, 2015 10:41 pm

I have the same question.
I hope to get an significant p value for G*E term. How to get that p value. And coefficients for that interaction term.
Thanks!

agiri_genepi
Posts: 2
Joined: Thu May 21, 2015 8:13 pm

Re: Understanding probabel GxE interaction output

Postby agiri_genepi » Wed Jun 17, 2015 8:46 pm

Hello,

You can obtain the p-value for the interaction term by dividing the SNPxcovariate_interaction_beta by the SNPxcovariate_interaction_standard_error. This gives you a z-score which you can convert into p-value using R doing something like this for example: 2*pnorm(-abs(z-score). This should give you a two-sided p-value for interaction.

My original question however has not been answered at all, which is: What is the chi2_SNP test statistic for (in the event we run an interaction analysis):

a) the SNP
b) the interaction term of the SNP and the covariate
c) the whole model in question compared to the null-model.

Help with this query would be greatly appreciated.

Thank you much.

nfishbane
Posts: 3
Joined: Wed Jun 03, 2015 7:47 pm

Re: Understanding probabel GxE interaction output

Postby nfishbane » Mon Jul 20, 2015 8:28 pm

I have verified the ProbABEL results manually in R using the imputed dosages. It seems as though when running palinear with an interaction term (GxE), the chi2_SNP test statistic is a 2-df statistic from Likelihood Ratio Test. The full model is containing all 3 terms G, E, and GxE (plus the covariates), and the null model only contains E (with the same covariates). So it considers the "null model" to be without both genetic terms (G and GxE). I'll denote this as the Likelihood Ratio L(GxE)/L(E) which asymptotically approaches a 2-df chi-square statistic.

I think if we want P-values for the individual terms, we will need to perform either
1. A Wald Test using the coefficient and the SE
2. Another ProbABEL run without the GxE term to obtain a chi2_SNP statistic representing the 1-df L(G,E)/L(E) statistic. This way we can obtain, say, L(GxE) /L(G,E) chisq-statistic which is also 1-df, for a test of the interaction term by itself.

Hope this is helpful!

lckarssen
Site Admin
Site Admin
Posts: 321
Joined: Tue Jan 04, 2011 3:04 pm
Location: Utrecht, The Netherlands

Re: Understanding probabel GxE interaction output

Postby lckarssen » Wed Jul 22, 2015 12:30 pm

Thanks Nick! I hadn't had the time yet to check the code before answering.

I'll update the manual to make this more explicit.
-------
Lennart Karssen
PolyOmica
The Netherlands
-------


Return to “ProbABEL”

Who is online

Users browsing this forum: No registered users and 2 guests