[BioC] GOstat

Seth Falcon sfalcon at fhcrc.org
Fri Dec 1 22:03:27 CET 2006

"James W. MacDonald" <jmacdon at med.umich.edu> writes:
> Consider the following code:
> library(hgu133plus2)
> library(GOstats)
> library(Rgraphviz)
> lls <- unlist(as.list(hgu133plus2ENTREZID))[1:500]
> lls <- lls[!is.na(lls)]
> univ <- unlist(as.list(hgu133plus2ENTREZID))
> univ <- univ[!is.na(univ)]
> params <- new("GOHyperGParams", geneIds=lls, universeGeneIds=univ,
> ontology="BP", annotation="hgu133plus2", conditional=TRUE)
> hypt <- hyperGTest(params)
> grph1 <- getGoGraph(params, summary(hypt)[1:3,1])
> plot(grph1)
> grph2 <- GOGraph(summary(hypt)[1:3,1], GOBPPARENTS)
> plot(grph2)
> Is the intent that the two graphs should be identical? My reading of
> the help page leads me to believe so.

I goofed this one up.  getGoGraph is used internally to generate the
graph used in the computation of the hyperG result.  For this purpose,
it is more convenient to have the graph with edges pointing to _more_
specific nodes (children) instead of the standard tree view which goes
the other way.

The method should never have been exported in the first place.  Even
if I change it to use GOXXPARENTS, all it does is pull out the
ontology slot and call GOGraph.  It hardly seems worth having, IMO.

Thanks for bringing this to my attention.  My plan for getGoGraph is:

  i) make it behave like GOGraph using GOXXPARENTS
  ii) deprecate it and make it go away asap.

I'd rather focus on some useful graph generating methods for the
result object.

Best Wishes,

+ seth

More information about the Bioconductor mailing list