[BioC] Wilcoxon rank sum test on Golub dataset

Sean Davis sdavis2 at mail.nih.gov
Thu Apr 20 12:53:45 CEST 2006

On 4/20/06 5:52 AM, "Andrej Kastrin" <andrej.kastrin at siol.net> wrote:

> Dear BioC useRs,
> I'm working on classical Golub dataset and I would like to select 10
> genes that are mostly overexpressed in AML, and 10 genes that are mostly
> overexpressed in ALL by using Wilcoxon rank sums test.
> I try with the below code (I paste just the core of the loop) which
> compute p value for each row, but the result is identical:
> wilcox.AML.pvals[i]  <- wilcox.test(aml.i,all.i)$p.value
> wilcox.ALL.pvals[i]  <- wilcox.test(all.i,aml.i)$p.value


Doing a wilcox.test just looks for differences between the two groups.  With
your best p-valued genes (using either list--they should be identical), you
can look at the magnitude of difference in means between AML and ALL and
choose genes that meet your criteria.  In other words, look for the ten
highest genes in your p-value list that have a mean in ALL that is higher
than in AML, for example.

However, this isn't really the way that most people think of
differentially-expressed genes.  Do you have a reason to want to look at 10
genes only?


More information about the Bioconductor mailing list