[BioC] GOStats: HyperGTest question
Seth Falcon
sfalcon at fhcrc.org
Thu Mar 8 19:36:05 CET 2007
Hi,
Johannes Graumann <johannes_graumann at web.de> writes:
>> params <- new(
>> "GOHyperGParams",
>> geneIds=ipilist,
>> universeGeneIds=names(as.list(get(paste(annotation,"GO",sep="")))),
>> annotation=annotation,
>> ontology="BP",
>> pvalueCutoff=0.001,
>> conditional=FALSE,
>> testDirection = "over"
>> )
>> hgOver <- hyperGTest(params)
>
> with 'ipilist' being a vector of IPI ids, and 'annotation' being a
> AnnBuilder package I build to map go terms to IPI ids via EntrezIDs,
> I get the following error:
>
>> Error in getUniverseHelper(probes, lib, entrezIds) :
>> No Entrez Gene ids left in universe
>
> The 'universe' construct is supposed to return all IPI ids from the
> annotation package.
>
> Please enlighten me as to where I err!
The environments from the annotation data packages that get used for
the calculation are hard-coded. Except when using the YEAST package,
the geneIds should be Entrez Gene IDs or at least be the same as what
is in the yourpkgENTREZID environment.
It isn't clear to me what you have and your your IPI IDs are.
> Something else: "GOHyperGParams" is not well documented on my system:
>
>> ?GOHyperGParams-class
Please try:
?"GOHyperGParams-class"
or
class ? GOHyperGParams
> My specific questions are:
> - is there an option like "ontology='all'"?
There is not currently such an option. You have to test each ontology
seprately.
I believe that the current approach is more conservative. The
universe of Entrez IDs is reduced to those that have at least one
annotation in the specified GO ontology. If one tested all ontologies
together, the universe would grow and, in general, you would get
smaller p-values for the same gene list.
> - is there an option like "testDirection='both'"?
Sorry, no. Are you thinking about something that is combined?
Computing both results isn't difficult:
pUnder <- pOver
testDirection(pUnder) <- "under"
ansOver = hyperGTest(pOver)
ansUnder = hyperGTest(pUnder)
+ seth
--
Seth Falcon | Computational Biology | Fred Hutchinson Cancer Research Center
http://bioconductor.org
More information about the Bioconductor
mailing list