[BioC] Re: remlscore function / random effects

[Gordon Smyth]

Hi Francois,

When there the mixed model contains only one random effect, apart from the 
residual, I use the function randomizedBlock() from the statmod package 
which is quite a bit faster than lme() and also (I think) more likely to 
converge successfully. When there are two or more random effects, I don't 
know of any alternatives in R to lme(). Doug has promised a faster version 
of lme() as part of a large revision of the nlme package, but I don't know 
when that will be available.

Unfortunately, remlscore() fits a very specialized type of model. It 
estimates a variance function rather than random effects and so isn't an 
alternative for this problem.

Best regards
Gordon

At 03:25 AM 10/02/2004, Francois Collin wrote:
>Hi Gordon,
>I'd like to fit a mixed effect model to expression
>values for each probe set on the U133A microarray.
>Many on the BioConductor list have the same problem.
>One solution is to call lme() within a loop (apply(),
>or for()).  I would have thought that using the
>remlscore() function would provide a more efficient
>alternative (in part because we can specify the X and
>Y matrices once outside the loop, maybe other
>reasons).  I intend to make the test myself, but if
>you know of a reason why this is not going to work I'd
>appreciate the tip.
>Thanks.
>
>-francois
>
>ps. If this question deserves an answer that would
>benefit others, feel free to post it on the BioC list.