Michael Breen | 28 Jun 18:01 2013
Picon

Re: edgeR multiple contrasts Vs. One test

Great to know!

So the FDR does scale across multiple contrasts.

Thank you!

On Thu, Jun 27, 2013 at 4:40 PM, Gordon K Smyth <smyth@...> wrote:

> Dear Michael,
>
> If I understand you correctly, you are asking about adjusting the p-values
> for multiple testing.
>
> The default in edgeR is to adjust the p-value in order to control the
> false discovery rate (FDR).  If you control the FDR at a given level for
> each of 5 contrasts separately, then you have automatically controlled the
> FDR at the same level for all 5 contrasts together.  The FDR is a scalable
> quantity in this sense.
>
> The situation would be different if you used adjust.method="holm". Holm's
> method controls the family-wise type I error rate, and the type I error
> rate does not scale over multiple contrasts.
>
> Best wishes
> Gordon
>
>  Date: Wed, 26 Jun 2013 12:44:14 -0700
>> From: Michael Breen <breenbioinformatics@...**≥
>> To: bioconductor@...
>> Subject: [BioC] edgeR multiple contrasts Vs. One test
>>
>> Hi All,
>>
>> If we have an design for which we have 4 groups, lets call:
>>
>> 1.Control Untreated
>> 2. Control Treated
>> 3. Cases Untreated
>> 4. Cases Treated.
>>
>> and we were interested in differences between:
>> -treated and untreated for Control
>> -treated and untreated for Cases
>> -treated differences between cases and controls
>> -untreated differences between cases and controls.
>> -differences between treated and untreated.
>>
>> 5 tests in total. We can then use edgeR contrast function as something
>> like this...
>>
>> contrasts <- makeContrasts(
>> Case.TreatedvsUntreated = Case.Treated-Case.Untreated,
>> Control.TreatedvsUntreated = Control.Treated-Control.**Untreated,
>> CasevsControl.Untreated = Case.Untreated-Control.**Untreated,
>> etc..... levels=design)
>>
>> This produces an appropriate rank order of significance for each
>> contrast. However, what is the cost of having no correction for the fact
>> that I just performed 5 tests on each gene instead of just 1 test??
>>
>> Any insight?
>>
>> Yours,
>>
>> Michael
>>
>
> ______________________________**______________________________**__________
> The information in this email is confidential and inte...{{dropped:10}}

_______________________________________________
Bioconductor mailing list
Bioconductor@...
https://stat.ethz.ch/mailman/listinfo/bioconductor
Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor


Gmane