[BioC] adjusted p values

[Gordon Smyth]

At 01:01 PM 19/06/2007, Naomi Altman wrote:
>I am not completely sure about 2, but you have interpreted 1 
>correctly.  I don't think we should call these "adjusted p-values" 
>because they are not, but this terminology seems to be used.
>
>--Naomi
>
>At 09:15 PM 6/18/2007, Lev Soinov wrote:
>>Dear List,
>>
>>   I've got a bit confused reading previous discussions from the 
>> list on the terminology about q values, adjusted p values and FDR 
>> in the LIMMA. Please, could you correct me if I am wrong here:
>>   1. BH adjusted p values from topTable or from decideTests are 
>> actually FDR. So, if we select genes with adjusted p values <0.05 
>> then we estimate that 5% of the selected genes are not really 
>> differentially expressed, i.e. false positives.
>>   2. when using decideTests with method="nestedF", p.value =0.01 
>> and adjust.method="BH", does it give you the list of genes for 
>> which BH adjusted F.p values <0.01, so one should expect 1% of 
>> false positive results in the list of the selected genes?

Yes, except that you'd expect less than rather than equal to 1% false 
discoveries, because 0.01 is a bound not an estimate.

And note this applies only to the F-stat list of genes, not to the 
more smaller lists of genes you get for individual contrasts. As I've 
said before on the list, nestedF does not give formal FDR control at 
the contrast level.

Gordon


>>   With kind regards,
>>   Lev.