[R-sig-ME] covariance problems in MCMCglmm. genetic effects and multiple membership
Alexandre Martin
alexandre.m.martin at gmail.com
Wed Aug 12 15:03:46 CEST 2015
Dear all, dear Szymek,
Thank you for your answer.
I am looking at ASReml to manage that.
All the best.
Alexandre
De : Szymek Drobniak [mailto:szymek.drobniak at uj.edu.pl]
Envoyé : 10 août 2015 13:28
À : Alexandre Martin; r-sig-mixed-models at r-project.org
Objet : Re: covariance problems in MCMCglmm. genetic effects and multiple membership
Hi Unfortunately this is not possible as far as Know but please someone correct me if I'm wrong - MCMCglmm cannot use design matrices of necessary form for random effects. you'll have to use asreml - or asreml-r which has syntax similar to MCMCGLMM.
Cheers
Szymek
W dniu pon., 10 sie 2015 o 19:15 Alexandre Martin <alexandre.m.martin at gmail.com> napisał(a):
Dear Szymek,
Thank you again for your contribution with your last answer.
I am wondering now how to write the model to estimate the covariance between the “mm(…)” and the “animal” random parts.
Jarrod wrote in this thread (https://stat.ethz.ch/pipermail/r-sig-mixed-models/2011q4/017034.html) about maternal genetic effects : “it is not yet possible to fit the covariance between maternal genetic and direct genetic effects“
Since my model is similar to that described in the thread, is it still impossible to fit a model to assess the covariance between direct and indirect genetic effects in a way similar to that in ASReml as follow?
For y =Xb + Z1animal + Z2maternal + e
Analysis of some kind
anim !P # The variable ‘anim’ is related to a pedigree fil
dam !P # The variable ‘dam’ is related to a pedigree file
dage 10 !A
rt 6
wwt
grp 322 !A
example.ped
example.dat
wwt ~ mu rt dage !r anim dam !f grp
0 0 1
anim 2
2 0 US !GP
.2 0 .15
anim o AINV
Best
Alexandre
De : Szymek Drobniak [mailto:szymek.drobniak at uj.edu.pl]
Envoyé : 14 mai 2015 17:39
À : r-sig-mixed-models at r-project.org
Cc : alexandre.m.martin at gmail.com
Objet : Re: genetic effects and multiple membership in MCMCglmm (Alexandre Martin)
Hello Alexandre,
using ped assigns a given correlation structure only to the "animal" random effect. You have to create an inverse of A matrix:
my_inverse <- inverseA(ped)$Ainv
and assign it in MCMCglmm to specific random effects:
MCMCglmm(Y~1,
random=~animal+mm(m1+m2+m3),
ginverse=list(animal=my_inverse, m1=my_inverse, m2=my_inverse,m3=my_inverse), data=dat, pr=T).
Cheers
Szymek
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list