[BioC] Limma background subtraction etc

Gordon Smyth smyth at wehi.edu.au
Thu Oct 2 19:50:16 MEST 2003


At 01:54 AM 2/10/2003, Jason Skelton wrote:
>Hi Gordon and everyone
>
>I'm having difficulty running the background correction function in LIMMA
>I'm trying to use the "minimum" option but get the following error message:
>
> > AR123rep6RGback <- backgroundCorrect(AR123rep6RG, method="minimum")
>Error in if (any(i)) { : missing value where TRUE/FALSE needed
>
>options "none" & "half", appear to work perfectly fine, any ideas ?

My guess is that there are missing values in your foreground or background 
intensities. Assuming that is the problem, I have implemented a fix in 
limma 1.2.7.

>once I have "corrected" I take it this just creates another RG object ?
>which can then be used in normalizeWithinArrays/normalizeBetweenArrays ?

Yes.

>ALSO
>
>Has anyone used the QVALUE function from John Storey with limma ?
>are then any functions in limma that support the use of QVALUE
>if not do you have any suggestions as to the best way to use it with limma.

No. John Storey's qvalue() function is copyrighted by him and so cannot be 
distributed with Bioconductor. In any case, it wouldn't change the limma 
results. The limma functions are designed to give a best ranking of the 
genes in order of evidence for differential expression. Although p-values 
are output, they are intended for ranking genes and should not be trusted 
as absolute measures of significance because they rely on assumptions of 
independence and normality which are surely not true (as do all similar 
methods). The qvalue() function won't change the results because it doesn't 
change the ranking of the genes. It's simply a more sophisticated version 
of the adjust="fdr" option already provided in toptable.

>If it is somehow possible to add another column of data to the toptable 
>command to display M, t, P.values, B and q.values ???

You can always add another column yourself to the output from toptable 
since toptable simply produces a data frame.

BTW, my personal approach is to estimate actual false discovery rates from 
spike-in control data rather than to try to estimate them from p-values in 
the absence of knowledge of the independence between genes.

Gordon

>thanks for any advice
>
>Jason
>
>--
>--------------------------------
>Jason Skelton
>Pathogen Microarrays
>Wellcome Trust Sanger Institute
>Hinxton
>Cambridge
>CB10 1SA
>
>Tel +44(0)1223 834244 Ext 7123
>Fax +44(0)1223 494919



More information about the Bioconductor mailing list