[R] Mixed effects model with a phylogenetic tree/ distance, matrix as a random effect

Terry Therneau therneau at mayo.edu
Tue Jun 4 15:51:54 CEST 2013


Take a look at lmekin() in the coxme package.  The motivating data set for my development 
of coxme was the Minnesota Family Breast Cancer project: 24050 subjects in 462 families.  
The random effect is an intercept per subject with sigma^2 K as its variance where K is 
the kinship matrix (1 for self-self, .5 for parent-child or sib-sib, .25 for uncle-neice, 
etc).  lmekin is a linear models front end to the same underlying routines.

    I think you want lmekin(y ~ x1 + x2 + (1| subject), data=yourdata, varlist= D)
or some such, where D is the similarity or "correlation" form of you distance matrix.

    A downside is that lmekin is sort of the poor cousin to comxe -- with finite time I've 
never gotton around to writing predict, residuals, plot, ... methods for it.  The basic 
fit is fine though.

Terry Therneau

(In general I agree with Bert & Ben to try the other list, but I don't happen to read it.)

On 06/04/2013 05:00 AM, r-help-request at r-project.org wrote:
> Hi,
> I'm trying to build a mixed-effects model in which I'd like to include
> either a distance matrix or a phylogenetic tree as a random effect.
> The troubles I've had are that:
> 1. Function lmer() in package lme4 only accepts a data frame column as a
> random factor and not a distance matrix.
> 2. Function MCMCglmm() in package MCMCglmm only accepts a rooted and
> ultrametric phylogenetic tree as a pedigree argument while my tree is
> neither (and for various reasons I cannot construct one or coerce mine
> to be a rooted, ultrametric tree).
>
> Is there any way around it?
> I'd appreciate mostly a solution to problem 1.
>
> Roey



More information about the R-help mailing list