[BioC] limma: decideTests, which option to choose and nestedF pvalues

[noel0925 at sbcglobal.net]

Hi All, 
This message can be ignored- I found my mistake.

Apologies,
Noelle

--- "noel0925 at sbcglobal.net" <noel0925 at sbcglobal.net>
wrote:

> Hi All,
> 
> I am trying to grasp the different options for
> decideTests in Limma. I have read
> many of the postings, but still am confused. Some of
> these postings left me with
> the following questions.
> 
> For the decideTests function in Limma, the method,
> "separate" is considered less stringent than the
> others. Does this mean least "stringent" in terms of
> identifying DE genes?  
> So is this method also considered to be the most
> conservative- eg it identifies
> fewer gene and also fewer false positives? Or is
> that a separate issue? I plan
> to use BH's FDR correction at say the 0.05 level so
> I realize then that within
> my list of DE genes, the expected number of FPs is
> less than 5%. So, this is
> only indirectly related to the method- separate,
> hierarchical, global, or nestedF?
> 
> This posting:
>
https://stat.ethz.ch/pipermail/bioconductor/2005-July/009542.html
> also states that "for example, the nestedF method is
> most
> powerful for picking up genes which change in
> multiple conditions, but is
> possibly least powerful for picking up genes which
> are different in only one
> condition". So then if you expect that some genes
> will change in many conditions
> (contrasts of interest) but that certain genes will
> be altered only in a certain
> condition, which approach do you choose?
> Is the "separate" approach the safest bet?
> 
> 
> Also as regards a separate but related posting:
>
https://stat.ethz.ch/pipermail/bioconductor/2006-February/011970.html
> about obtaining individual p-values corresponding to
> the nestedF method.
> Here the post reads:
> "No, there is no way to get individual p-values
> corresponding to the nestedF
> method.  You can however just used the overall
> F-test p-values, fit$F.p.value,
> which will give you p-values at the probe level
> rather than the contrast level.
> Use p.adjust(fit$F.p.value) to get adjusted
> p-values."
> 
> Isn't it possible to obtain the individual p-values
> corresponding
> to the nestedF method using the following:
> 
>
NestedF<-decideTests(fit2,method="nestedF",adjust.method="BH",p.value=0.05)
> write.fit(fit2, results=NestedF,
> "NestedF_FDRadjusted_P0.05.txt", adjust="fdr",
> sep="\t")
> 
> Here I obtain a table with p.values for each of my
> contrasts and a separate one
> for the overall F-test. These values differ from
> that obtained from say topTable
> which is eqivalent to performing decideTests with
> method="separate" 
> when the same multiple testing correction is used,
> namely "BH" with p=0.05.
> 
> So, if these are not the individual p-values from
> the nestedF method, then what
> do they represent?
> 
> Thanks in advance, 
> Noelle
> 
>