Logistic Regression With Recessive Model

Question

[Hamdi]

Hello, I would like tu run analysis using PLINK for SNPs but with recessive model as the default model used in logistic test is additive. could you please give me your advice. thanks a lot in advance PS: I tested the SNP using logistic regression test and 6 covariates but this time the reviewer asked for the same test but in recessive model instead of the default additive model

Answer 1

[Thomas]

At the plink homepage there is a very nice and clear guideline for performing different genetic models:

http://pngu.mgh.harvard.edu/~purcell/plink/anal.shtml#glm

From the homepage:
To specify a model assuming full dominance (or recessive) for the minor allele (i.e. rather than the 2 df model mentioned above), you can specify with either
--dominant
or
--recessive

regards
Thomas

[Hamdi]

thank you Thomas, I will perform this analysis as suggested regards

Answer 2

[Genotepes]

Hi,

the way of doing it is pretty well explained in plink web page (--logistic --recessive). However, the other question is whether it is wise to do so.

What will be the genotype frequency of you rare genotype under the recessive model. Especially as you are also including six covariates (are you sure they are all independent - at least a little bit).

I think that if you really need to get rid of a reviewer, you may also want to test the full model expressed with additive and dominance component, in order to see whether this dominance component is significant (showing deviation from additive).

Christian

[Hamdi]

thanks Christian for your valuable comment. However the most significant ORs were detected among individuals carrying two minor alleles and in the 1st stage analyis our collaborators used pseudomarker which incorporate recessive and dominant models, thus we need to have the same genetic model tested. regards Hamdi

[Genotepes]

I understand. Then just --logistic --recessive. I don't remember whether you can use permutation in such a complex setting (covariate with logistic and linear) --mperm 100000 for instance. Should be better to do it.