observation level random effects/kinship model

When I was saying that ASREML was the only package allowing something, it
was not association mapping but the possibility to specify a
variance/covariance matrix by directly importing the entire matrix (as in
the example of specifying an external kinship matrix for a genetic effect).

On Wed, Mar 14, 2012 at 8:28 AM, Ryan King <c.ryan.king at gmail.com> wrote:

You can also use MCMCglmm and R-INLA.

Ryan, have you got some links for what you talk about ? I'm quite surprised
that there are already packages for doing that because a lot of geneticists
are using ASREML even if they have to pay for it. I really hope that there
is a good reason for that !

ASREML is pretty quick and robust for large problems.

I am guessing you want to specify a large (nonsparse) empirical kinship 
matrix.  Then lmekin, in the kinship package, is one R package that allows 
you to do this, but it gets slow for large datasets.  I have hypothesized, 
but never got round to trying, that coxme() in the same package could be 
abused to give a binomial GLMM ;).  AnimalINLA allows one to fit 
arbitrary matrices too:

  If not using compute.Ainverse to calculate the precision matrix
  [the inverse relationship matrix], the precision matrix has to be
  on the form sparseMatrix(i = ,j = , x =), the two first (i ,j)
  are the individuals compared in the relationship matrix (remember
  the individual numbers must match in the relationship matrix and
  the individual number in data (genetic)), third list element
  (values) are the precision values (the corresponding element of the
  precision matrix).


The regress package does gaussian mixed models only.

| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v