[BioC] fdr cutoff

[Richard Friedman]

On Oct 1, 2013, at 3:16 AM, Smeeta shrestha wrote:

> Dear Dr Friedman,
> 
> I am keen to know the cut off for Adjusted- pvalue . I saw your post to Delasa.(I have attached the whole post in this email for your reference)
> 
> "Say you use a p-value cutoff <0.05. As long as this value is not exceeded it is reasonable to take p-adjust (fdr) at 0.1 or even as high as 0.5 as long as you know that in the latter case have the genes you try to validate by pcr are expected not to be differentially expressed. I think that p-adjust<0.05-0.25 are good cutoffs".
> 
> I would like to know is there any material which explains or justifies the reason behide how much fdr cut off I can take.
> 
> My data with p value 0.05 shows a huge set of genes. But when corrected using Benjamini Hochberg correction. I get no genes on adjusted pvalue cut off of 0.05. The closest i get is 0.556.
> 
> Please any input would be most helpful.
> 
> Thank you for your attention.
> 
> sincerely
> Smeeta Shrestha
> Post doctoral Fellow
> CCMB
> Hyderabad
> India.
> <fdr.docx>


Dear Smeeta,

	First of all, it is best to copy the list in all queries.
I know of know specific reference and indeed I ventured
above fdr<0.05 based on a conversation with
Terry Speed (and I hope that I am not misquoting or 
misinterpreting)  in which he said that the fdr cutoff
"depends upon what you use the data for". This makes sense, in that
the fdr is the number of statistically significant genes
by the rawp<0.05 criterion that are likely to be false.
So fdr = 0.556 means that the Benjamini-Hochberg
fdr predicts that there is a 56% chance that your most
signficant gene really is not significant. So it will be a bad
bet to expect this gene to be validated by pcr,

	I would be interested in what other correspondents
have to say in this matter.

Best wishes,
Rich

Richard A. Friedman, PhD
Associate Research Scientist,
Biomedical Informatics Shared Resource
Herbert Irving Comprehensive Cancer Center (HICCC)
Lecturer,
Department of Biomedical Informatics (DBMI)
Educational Coordinator,
Center for Computational Biology and Bioinformatics (C2B2)/
National Center for Multiscale Analysis of Genomic Networks (MAGNet)/
Columbia Department of Systems Biology
Room 824
Irving Cancer Research Center
Columbia University
1130 St. Nicholas Ave
New York, NY 10032
(212)851-4765 (voice)
friedman at cancercenter.columbia.edu
http://cancercenter.columbia.edu/~friedman/

"You shouldn't have told Isaac to
say 'hi' to Snowden when he's at
the Moscow airport. He will  do it
and then he won't be able to get
a computer job."

- Judith Friedman, 
Age withheld on request