[BioC] limma(B) and maanova(La.svd) questions
laugraud at biomserv.univ-lyon1.fr
Thu Oct 20 15:54:08 CEST 2005
I can only agree with you on the quality of the data (it was a 2 arrays
dye swap test so that's why there is no replicate), it could be a very
I can manage to figure out this problem linked with LIMMA.
However I can't understand the link with what I have seen in maanova and
the problem I met.
I would still be very interested in knowing if people have had the same
kind of things.
Morten Mattingsdal wrote:
> Basically you have a weak experiment with too few gene observations
> (only 2 arrays if I understood you right).
> With so few observations you expect to get rather poor statistics.
> I would expand your experiment (dont dye-swap:) with more biological
> replicates if I were you.
> Aurélie Laugraud wrote:
>> I am working on 2-color arrays, 2 arrays (2 conditions), 27648 spots.
>> my first try with maanova was very unsuccessful as
>> there was an problem with the degrees of freedom ( model contained
>> Array+Dye+Sample ...I don't think it is too much but it was). My
>> second try was to put Array as random but I got this:
>> "Fitting mixed effect model...
>> Erreur dans La.svd(X, method = "dgesvd") : infinite or missing values
>> in 'x'"
>> I have no idea what to change to make it happy, I however found out
>> that it is very likely the function "pinv" in "mixed" that is unhappy...
>> Then I tried limma, everything went find and ok until I had a look at
>> the "B" value which was terrifyingly low (the best I can get is 0.67
>> for the best score). I tried quite a lot of different values for the
>> parameters but it doesn't seem to
>> change the real underlying problem there.
>> Does anyone haven any idea or have seen the same things ?
>> Bioconductor mailing list
>> Bioconductor at stat.math.ethz.ch
PRABI - LBBE
Bât. G.Mendel- 1er étage
16, rue Raphaël Dubois
69622 Villeurbanne Cedex
More information about the Bioconductor