[BioC] Counterintuitive results from edgeR

Gordon K Smyth smyth at wehi.EDU.AU
Sat Apr 19 02:06:15 CEST 2014

Dear Son,

I am not seeing anything counterintuitive here.

edgeR gives you a P-value of 0.02 for this gene, which agrees perfectly 
with your intuition that the gene looks differentially expressed if 
considered in isolation.

The logFC is reported to be 0.99, which looks pretty similar to the change 
in average logCPM from group A to group B.  This all seems as expected.

The reason for FDR=1 is that, given the number of genes in your experiment 
and given the variability in your experiment, you could easily fish around 
in the data to find a gene that looks as DE as this one, even if there 
were in fact no truly DE genes.  This is the difference between testing 
for DE for one gene in isolation, versus adjusting for multiple testing 
over thousands of genes.

In general, you should not expect a gene with a modest p-value like 0.02 
to be chosen as significantly DE after adjusting for multiple testing over 
tens of thousands of genes.

Best wishes

> Son Pham kspham at eng.ucsd.edu
> Sat Apr 19 01:01:39 CEST 2014
> Hi edgeR community:

> I'm testing for DE genes in two groups. Group A has 6 replicates, group 
> B has 4 replicates. I spot a very strange gene, that should be DE but 
> somehow edgeR doesn't report it.
> Here is the log2CPM of group A:
> (5.8804275247607; 5.95666869234362;  6.24683910018322; 4.03634160591149;
> 4.9390167413036;  5.22009594582392)
> and log2CPM of group B:
> (6.68129783945799;   6.31414609301072;       6.94357946236393;
> 6.19026250939524)
> It's clear for us that these two groups has different CPM: B > A.
> But here is how edgeR reports:
> MyGeneID  0.990998868329884(logFC)       6.11753790162748(LogCPM)
> 0.0238675732815782(Pvalue)      1(FDR).
> This is very counter intuitive. Why here we have FDR = 1 ? and even I 
> can not infer how logFC = 0.99
> Any clarification is greatly appreciated !
> Thank you,
> Son.

The information in this email is confidential and intend...{{dropped:4}}

More information about the Bioconductor mailing list