[GenABEL-dev] faster polygenic

[Xia Shen]

Hi Yurii,

I just 'sudo-randomly' tried some different argument value in the hglm function call, and happened to have solved the bias issue (at least it seems to be so). Please try putting the following arguments in your hglm wrapper:

hglm(y = ph, X = desmat, Z = L, conv = 1e-6, maxit = 40)

I re-did 100 simulations and the bias is not sig from 0 with p-value = 0.77!

I guess the problem here is just numerical. Since IBS matrix is generally sparse (close to identity), the estimation of variance components will be sensitive numerically. Just make the convergence criteria more strict, and it works.

Note that this might slow down hglm a bit (not very much I guess), but I think it should be more than 20x faster than the old polygenic. Not familiar with fmm, but: 1. our hglm can deal with any GLM family of the trait distribution, even non-normal genetic effects. 2. we produce standard errors for everything - every part is checkable using h-likelihood :)

Cheers

Xia


On Feb 28, 2011, at 12:24 AM, Yurii Aulchenko wrote:

> some extra info:
> 
> was running the checks cf. simulated heritability
> 
> t-test
> 
> difference (h2simulated - h2estimated-old-polygenic) ns from 0, p=0.36
> 
> difference (h2simulated - h2estimated-hglm) SIG from 0, p=0.006
> 
> difference (h2simulated - h2estimated-fmm) ns from 0, p=0.30
> 
> So, my feeling is that hglm does something wrong; old and fmm do very
> well, but old does better :)
> 
> On Sun, Feb 27, 2011 at 10:51 PM, Yurii Aulchenko
> <yurii.aulchenko at gmail.com> wrote:
>> Dear All,
>> 
>> Some of you may know, that MixABEL::FastMixedModel by William Astle
>> first does polygenic model estimation (e.g. h2-estimate is given by
>> FastMixedModel(...)$null.herit), though it is build 'within' GWA and
>> it is not yet possible to get hold on all necessary 'null'
>> heritability estimates.
>> 
> _______________________________________________
> 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