[BioC] Significancy of not differentially expressed genes

[Sean Davis]

On 5/11/06 3:53 AM, "Mohammad Esad-Djou" <shahrgol at web.de> wrote:

> Hello Naomi, 
> 
> now I have a question.  :)
> 
>>> I am looking for a measure of the significance for the potential
>>> 'not differentially expressed' call that could be given to a gene
>>> and that can be considered reliable from the community.
> 
> I believe that you can find with multiple Testing (e.g. SAM, siggenes) 'not
> differentially expressed'  genes.
> If H0 is correct: genes are not differentially expressed.
> did I understand correctly?
>  
> I am not sure, but I believe that multiple Testing is helpful.

Unfortunately, it is not quite that simple.  Hypothesis testing (in the
statistical sense) does not allow one to say that "if we have no evidence
for differential expression, the genes are expressed similarly".  It simply
means that we do not have evidence that the genes are differentially
expressed, which could happen for many reasons besides them not being
differentially expressed.  In other words, typical hypothesis testing lets
us guard against particular types of errors, and drawing conclusions based
on "accepting the null hypothesis" like you suggest does not allow one to
guard against those errors.

Perhaps if you have a particular situation that you are thinking of, you
could share it in case others on the list have any creative answers for you.

Sean