8.4 years ago by
Zentrum für Molekularbiologie, Universität Heidelberg
On 2011-06-13 14:56, Dawei Li wrote:
> Dear Simon Anders:
> I am using your DESeq package for data analysis.
> I have one little quetion regarding the package:
> In the manual, it is said that padj controls the FDR value. Can I
> call padj as FDR?
> My clients want the FDR value. I tell them padj is the FDR value.
The 'padj' values is what your clients want, however they and you are
incorrect in calling them FDR values.
An individual test (i.e., a gene) never has an FDR value, it has an
adjusted p-values. (Exception: so-called local FDRs, which I am not
The definition of a "p value adjusted for FDR control" is: If one
all tests (genes) with an adjusted p value below a threshold f, then
list of these genes contains (in expectation) at most a proportion f
false dicoveries, i.e., thresholding adjusted p values at f controls
false dicovery rate (FDR) at f.
Hence, the term "FDR" refers to the threshold you cut the adjusted p
values at, not to the adjusted p values themselves.