[BioC] Limma - small bug in decideTests?

Misha Kapushesky ostolop at ebi.ac.uk
Sun Jun 17 12:17:06 CEST 2007

Dear List,

I looked through the archives and didn't notice anyone pointing this out, 
so I thought I'd ask:

If decideTests() is called with the method argument set to hierarchical 
(by the way, that word is misspelled in the implementation as 
'heirarchical' - if someone should call it with correct spelling, they get 
an error), then internally classifyTestsP() is called to adjust across 
contrasts (after adjusting across genes). So far so good (exc the 

However, it seems the multiple adjustment method is not propagated to 
classifyTestsP - so if the user requests adjust.method="BH", 
classifyTestsP will perform the default adjustment, "holm", which is more 
conservative. Is that behavior by design? Or is there just a 
"method=adjust.method" missing on the classifyTestsP() call?

Also one could do fewer computations in the code, perhaps, if one did 
something like

   padj = p.adjust(object$F.p.value, method=adjust.method)
   sel = padj < 0.05
   pmax=min(padj[!sel], na.rm=TRUE)
   results=classifyTestsP(object[sel,], p.value=pmax, method=adjust.method)

for the hierarchical and nestedF branches of decideTests() - this avoids 
computing i, n, and a there. Or are these computed there for a reason, 


--Misha K.

More information about the Bioconductor mailing list