Dear Simon,
From my perspective, the idea behind developing a geneSet class is not
to store results from a GSEA run or from a typical class comparison but
to standardize the use of collection of genes having something in common
and allow for reproducible research.
As you know for GSEA to work the user has to specify (besides the
expression matrix and class labels) which is the set of genes that has
to be tested. One may want to test the set of genes belonging to a
pathway, or belonging to a GO term, or simply his own collection of
genes. The gene set(s) is just an argument to be passed somehow to GSEA
algorithm. The result of the GSEA test on the same gene set A=c("gene1",
"gene2"....) will differ function of many things such as, the expression
data used, number of permutations, etc. and the directions of change
will depend on the expression data used. The results may not generalize
from one experiment to another, but if this happens, we know at least
that this is not due to using different genes in the set A.
If you used say a t-test to test your individual genes and you got a top
table in which the log-ratios M will have a particular meaning for you,
then you can get the directions of change very easily with something
like
myTopTable[geneSet$geneIDs,"M"] or similar.
Adi
Dear Adi,
I see your points but I disagree.
A gene list is always derived from somewhere. Thus, there are
particular
experiments associated with it. Under this particular experimental setting, there are stereotypes of UP and DOWN. With that being said, we can
also >>find
metadata associated with a gene list is important to keep.
As such, I think the directions make sense; although mathematically
the >>sign
can be flipped, biologists have predefined contexts of UP and DOWN.
Simon