polygenic function

[merina]

Hello,

Is polygenic function in GenABEL suitable for use when the random effects are correlated?
I have a phenotype in binary format. In the "trait.type" criteria two options are possible gaussian or binomial. I have used "binomial" but I don't know which link function is being used. Is there anyway to check the link function?

Thanks,

Merina

[Nicola Pirastu]

Hi Merina,

to answer your question I need to ask you one first: what exactly are you trying to do?

In fact if you are only trying to run a GWAS on CC samples you can use ProbABEL, instructions can be found here on paragraph 3.2:

http://www.genabel.org/sites/default/fi ... manual.pdf

However the family option in polygenic does not really work and should not be used, if you are trying to get the coefficients of the covariates of a logistic regression you should use poligenic_hglm instead.

Hope this helps you.

Best

Nicola

[Daniel E. Weeks]

This was written in June 2013:

However the family option in polygenic does not really work and should not be used, if you are trying to get the coefficients of the covariates of a logistic regression you should use poligenic_hglm instead.

Is this statement still true?

I think it probably is, as if I try polygenic with trait.type = "binomial" with a single binary trait and no covariates, while the estimate of the mean makes sense, the subsequent FGLS check fails.

If polygenic still only works with trait.type = "gaussian", then shouldn't the trait.type = "binomial" option be removed from the polygenic code?

Thank you,
Dan Weeks

[asha2015]

Hi

I am getting the following error:

Code: Select all

Error in nlm(llFUN, p = parsave, y =y,desmat = desmat, relmat =relmat,
:
non-finite value supplied by 'nlm'


When I run:

Code: Select all

model_AF <- polygenic(af ~ sex, kin=gkin, data, trait.type="binomial")


I double checked the phenotype file and it is in recommended order. My phenotype file has 0/1

Please help

[asha2015]

Hi

in the following code:
model_AF <- polygenic(af ~ sex, kin=gkin, data, trait.type="binomial")

Does the trait type change statistical function? I always run into an error when I choose"binomial" but "gaussian" seems to work.


Kind regards