[BioC] GOStats - HyperGTest - expert opinion needed about the approach selecting gene list
James W. MacDonald
jmacdon at med.umich.edu
Sun Jan 6 23:46:30 CET 2008
I'm not too keen on this idea, mainly because the assumption here is
that all up-regulated genes are somehow related and the down-regulated
genes are related, but there is no relation between the groups.
Let's say that there are 20 genes related to GO term XXX. Half of them
are up-regulated, and half are down-regulated (because the first 10 have
a negative effect on the expression of the second 10). In this scenario
it may be that you won't see any significance for this term, but you
might if you didn't separate the two groups.
Now this is a pretty stupidly simplistic scenario, which in fact helps
prove my point -- for most genes we still have very little information
about what processes affect transcription, so making the assumption that
genes with positive correlation are related and those with negative
correlation are not is a pretty bold assumption indeed.
Srinivas Iyyer wrote:
> Dear all,
> I am analyzing gene diff. expression data for a rat
> instead of asking which biological processes (BP) are
> enriched for my diff. expressed genes, i am interested
> in asking which BP are enriched for upregulated genes
> and which BP are enriched for downregulated genes.
> For doing this, i seperated the diff. expressed gene
> lists into two lists (up and down regulated) after
> filtering the list for p-val 0.001 (this p-val is
> obtained from a t-test).
> now I make all up-regulated genes as one 'test list'
> to see enrichment for BP using entire 'chip'(genes/BP
> on present on chip) as universe.
> Similarly, I check enrichement for all downregulated
> genes keeping chip as universe.
> My question is:
> 1. Is it okay to do this way instead of making all up
> and downregulated genes into a one big list.
> 2. If i do not see many differential expressed genes,
> by doing this way, will I be getting all false
> positive enriched terms.
> 3. Am I missing any key concept and ending up in a
> flawed analysis.
> any suggestions will help me in big way.
> Thank you
> Never miss a thing. Make Yahoo your home page.
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor
More information about the Bioconductor