False discovery rates

[TimL]

Hi all:

Realizing that the occurrence of false discovery rates is somewhat
dependent upon what you set your priors too, has anyone suggested what
an average false discovery rate for the Bayesian approach to haplotype
mapping might be? I've searched a bit and found references for
regression-based approaches, but not for Bayesian approaches....Any
references would be helpful...

Cheers!

Tim L.

[Matthew Stephens]

you can estimate your local FDR for each SNP association identified
by computing the posterior probabiltiy of association from the BF
(eg via posterior odds = prior odds times BF).

An example, if your prior odds are 1/10^4 (ie one in 10^4 SNPs are
assocaited with outcome)
and BF is 10^5 then posterior odds = 10/1, which gives local FDR of
1/11 for that SNP. (1/[1+10])

Of course, this FDR calculation assumes that the priors you used are "correct",
But any FDR calculation is going to have to make assumptions about
effect size and
proportion of null SNPs.

Matthew