[BioC] RE: Bioconductor Digest, Vol 8, Issue 15

[Douglas Bates]

Although you didn't say so I presume you were replying to the message
on time-course experiments. (You quoted 10 different messages from a
digest in your reply.)

Is there a reason for not using the lme function in the nlme package
to obtain the maximum likelihood or REML estimates for the
mixed-effects model?

Even though I am one of the authors of lme I am not being coy in
asking this.  I'm not sure exactly what the mixed-effects model would
be and it is possible that the model could be fit easily in SAS PROC
MIXED but not with lme.  If that is the case then we (Saikat DebRoy
and I) could take this into account as we redesign lme for R.

"Baker, Stephen" <Stephen.Baker at umassmed.edu> writes:

> Gary Churchill at the Jackson Labs in Maine has an R program on his
> website for performing mixed models ANOVA on microarray data.  The only
> problem with this is it uses least squares to fit the model (which would
> include a within-subjects factor for the time effect) and would requires
> that there are no missing data points and all subjects being measured at
> the same time points.  This is because the least squares solution
> involves inverting a matrix and missing data would make it not of full
> rank.
> 
> An alternative approach which wouldn't be done in R would be to use PROC
> MIXED in the SAS stats package.  This uses maximum likelihood to fit
> mixed models and works well.  If you really want to try to do it in R,
> Yudi Pawitan at Dept. of Stats at University of Cork in Ireland has a
> book and a set of R programs which would give you a leg up on it:
> 
> http://statistics.ucc.ie/staff/yudi/likelihood/index.htm

> Date: Wed, 8 Oct 2003 12:02:36 +0200 (CEST)
> From: edoardo missiaglia <edo_missiaglia at yahoo.it>
> Subject: [BioC] time-course experiments
> To: bioconductor at stat.math.ethz.ch
> Message-ID: <20031008100236.12628.qmail at web11701.mail.yahoo.com>
> Content-Type: text/plain; charset=iso-8859-1
> 
> Dear all,
> 
> I am now working on some time-course experiments and I
> have applied to them some classical statistic methods
> to identify genes that change their expression between
> time points. However I have read few papers (such as
> Peddada et al. Gene selection and clustering for
> time-course and dose-response microarray experiments
> using order-restricted inference; GUO, X et al
> Statistical significance analysis of longitudinal gene expression data;
> etc..) where they describe specific methods for the analysis of this
> type of data. Unfortunately my background (I am biologist) make
> difficult to transform the algorithms reported in these papers in
> something usable in R. In the same time, I could not find packages in
> bioconductor that face this kind of problems ( there is only GeneTS
> written by Korbinian Strimmer, that is useful in a cyclic time-course
> experiment). I was wondering if anybody has already developed a package
> or some functions usable in R specifically designed for time-course
> experiment that consider the particular structure of this data.
> Otherwise is there anybody interest in developing something from
> scratch? Thank you very much in advance for your help.
> 
> Best wishes,
> 
> edoardo

-- 
Douglas Bates                            bates at stat.wisc.edu
Statistics Department                    608/262-2598
University of Wisconsin - Madison        http://www.stat.wisc.edu/~bates/