On Mon, Mar 3, 2008 at 4:59 AM, Mayte Garcia Conesa
<mtconesa at="" cebas.csic.es=""> wrote:
> To the attention of Bioconductor Help System,
>
> I have a question about the application of FDR cut-off values to
gene expression data. When you are looking at different sets of data,
can you apply different FDR cut-off values? Sometimes in one
experiment you may find that applying an FDR<0.05 yields a very small
number of significant changes but if you accept an FDR slighlty higher
(<0.075) then you may get a few hundred significant changes. What is
the ideal FDR cut-off value? Can you use different FDR cut-off values
for different experimental data? For example, if you have two sets of
experimentla data:
>
> Experiment 1: 30 Differentially Expressed (DE) genes for an
FDR<0.05 or 456 DE genes for an FDR<0.1
> Experiment 2: 678 DE genes for an FDR<0.05 or 2398 DE genes for and
FDR<0.1
FDR stands for False Discovery Rate and applies to a list of genes,
not any particular gene. If one has a list of 100 genes and an FDR of
0.05 for that list, then 5 of those 100 genes are expected to be false
discoveries. In that sense, the FDR has a nice intuitive meaning.
> If you want to compare results from the two experiments, is it
correct to compare the results with FDR<0.1 from experiment 1 against
the results with an FDR<0.05 from experiment 2?
I'm not sure what you mean by "compare", but comparing overlap of the
top N genes from one experiment and the top M genes from another
experiment is only one of many way of comparing two experiments. It
might be better to think about comparing broad categories of genes for
enrichment in the two experiments, as another possibility.
Sean
