Dear Jarrod and esteemed mixed-modelers,
First, thank you again for your help on a previous post.
I am returning on the work on indirect genetic effects I had to set aside
last month and I am trying Jarrod's solution using the function "mm".
I am unfortunately unable to run a model that links mate_1, mate_2, ...,
mate_n to the A inverse.
I used this dummy (and maybe too simple) data set to test the model:
ped=data.frame(animal=letters[1:11] ,dam= c(NA, NA, NA, gl(2,4,labels =
c("a","c"))), sire=c(NA,NA,NA,rep("b",8)))
dat= data.frame(animal=letters[4:11], pen=c(1,1,2,2,2,1,1,2), Y=rnorm(n =
8,mean = 5,sd =
0.4),m1=c("e","d","g","f","f","d","d","f"),m2=c("i","i","h","h","g","e","e
","g"),m3=c("j","j","k","k","k","j","i","h")),
where "m" stands for "mate". I have verified that each column had the same
factor levels (following Jarrod's suggestion : factor(mate_1,
levels=all.ids))
Using this dummy dataset in this basic model, mtest <- MCMCglmm(Y~1,
random=~animal+mm(m1+m2+m3),pedigree=ped, data=dat, pr=T),
mm(m1+m2+m3):animal (and other crazy possibilities) led to this error :
"interactions not permitted in str and mm structures".
Do anyone know how to link the different mates to the Ainverse in order to
obtain blups of m1, m2, ..., m_n.
I would be very grateful for any hint and instruction.
Many thanks again for your help.
Alexandre
-----Message d'origine-----
De?: Jarrod Hadfield [mailto:j.hadfield at ed.ac.uk] Envoy??: 3 avril 2015
03:51 ??: Jarrod Hadfield Cc?: Alexandre Martin;
'r-sig-mixed-models at r-project.org'
Objet?: Re: [R-sig-ME] design matrices in MCMCglmm
Hi,
Sorry it should have been just:
random=~mm(mate_1+mate_2+...mate_n)
and it is the mate_1, mate_2, ... mate_n that need to be linked to the A
inverse.
Cheers,
Jarrod
Quoting Jarrod Hadfield <j.hadfield at ed.ac.uk> on Fri, 03 Apr 2015
08:16:02 +0100:
Hi,
Have columns mate_1, mate_2 ... mate_n where n is group size. In each
column have the identity of each cage mate (order does not matter).
Make sure each column has the same factor levels even if they don't
appear. For example,
factor(mate_1, levels=all.ids)
where all.ids are all possible cage mates. Then fit:
random=~mm(mate_1+mate_2+...mate_n):animal
where animal is linked to the pedigree through the ginverse argument.
Cheers,
Jarrod
Quoting Alexandre Martin <alexandre.m.martin at gmail.com> on Tue, 31 Mar
2015 10:07:34 -0400:
Hi Jarrod,
Thank you for your help.
My question is now extended to the subject of associative indirect
genetic effects.
For example in this data set :
id cage
a 1
b 2
c 1
d 1
e 2
f 2
g 2
cage is a grouping variable describing the composition of cages.
For instance, individuals a,c,d live in cage 1.
Design matrix Z_cage typically produced by MCMCglmm should be:
c1 c2
a 1 0
b 0 1
c 1 0
d 1 0
e 0 1
f 0 1
g 0 1
where phenotype of individuals {a, b, ..., g} are linked to cages 1
and
2.
Design matrix Z_mates, however, linking the phenotype of individual i
to its cage' mates is:
a b c d e f g
a 0 0 1 1 0 0 0
b 0 0 0 0 1 1 1
c 1 0 0 1 0 0 0
d 1 0 1 0 0 0 0
e 0 1 0 0 0 1 1
f 0 1 0 0 1 0 1
g 0 1 0 0 1 1 0
It is Z_cage that is given by default, whereas it is matrix Z_mates
that should be used to predict associative effects.
Is it possible to force MCMCglmm to work with Z_mates instead of
Z_cage?
Thanks again!
Alexandre
Le 2015-03-28 04:01, Jarrod Hadfield a ?crit :
Hi Alexandre,
The design matrices should be identical for both effects (z_{ij}=1
if the jth individual is the mother of individual i). The difference
is in the correlation structure of the random effects. For
environmental maternal effects they are assumed iid (i.e. an
identity matrix) but for the maternal genetic effects they are
assumed to be proportional to the A matrix. inverseA will return the
inverse of A if you pass it the pedigree. It is this inverse that is
required for forming the MME.
Cheers,
Jarrod
Quoting Alexandre Martin <alexandre.m.martin at gmail.com> on Fri, 27
Mar
2015 16:39:40 -0400:
Dear all,
I am working on estimating maternal effects (genetic and
environmental) with MCMCglmm that is new for me.
I am trying to apply to MCMCglmm what is shown in online Muir's
course notes made for SAS. Leanning on Henderson?s Mixed Model
Equation, these notes explain how to solve MME to predict random
effects ?by hand?.
Here is my concern:
I do not know how to extract the design matrices for a MCMCglmm
model, e.g. the relatedness matrix or the one for maternal genetic
effects. I want that to understand how the design matrices are
constructed by comparing them to what they are supposed to look
like. For instance, the design matrix for maternal genetic effects
should relate offspring to all the individuals that are in the
pedigree, whereas the design matrix for maternal environmental
effects should just relate offspring to their mothers. Does such a
difference exist when MCMCglmm constructs its design matrices? If
not, how to include such different matrices in models?
Any help will be greatly appreciated. Thank you!
Alexandre
[[alternative HTML version deleted]]