[BioC] Re: Bioconductor Digest, Vol 22, Issue 18

[Gordon K Smyth]

> Date: Fri, 17 Dec 2004 11:39:26 -0800 (PST)
> From: Saurin Jani <saurin_jani at yahoo.com>
> Subject: [BioC] Limma  FDR and p.value
> To: Bioconductor Bioconductor <bioconductor at stat.math.ethz.ch>
> Message-ID: <20041217193926.31486.qmail at web41113.mail.yahoo.com>
> Content-Type: text/plain; charset=us-ascii
>
> Hi,
>
> This might be simple question, but I am confused
> somehow..if anyone can guide me through this that
> would be great.
>
>
>> fit<- eBayes(fit);
>
>> results_fdr1<- classifyTestsP(fit, p.value=0.01,
>   method="fdr")
>
>> vennDiagram(results_fdr1,include = "both");
>   (I get very nice vennDiagram, my all contrasts have
>    nice numbers of genes)
>
>> coef1_Genes <- topTable(fit,coef = 1, adjust =
>   "fdr",number = 50);
>
> Above when I get coef1_Genes, which has adjusted
> P.value way higher than 0.01, sometimes its range from
> 0.5.....0.904 or something like that. so, how can I
> decide my cutoff for FDR? I saw nice numbers in my
> vennDiagram above so, i thought I have lots of genes
> which has < 0.01 FDR but it not when I checked in
> coef1_Genes.
>
> In above results coef1_Genes P.value = FDR cutoff
> right? There is no option in topTable for howmuch FDR
> level..!

I think you are assuming that classifyTestsP() adjusts for multiple testing across genes. 
Actually it adjusts only across contrasts, i.e., across rows rather than down columns.  As the
help says

  "The adjustment for multiple testing is across the contrasts
  rather than the more usual control across genes."

You might want to use decideTests() instead.  It does something closer to what you are probably
expecting.

Gordon

> I set p.value = 0.01 in classifyTestsP function so, I
> think classifyTestsP gives me FDR level < 0.01,
> because I am using method = "fdr", and those nice
> number I see in vennDiagram.......Is this right
> understanding? if yes then how can I access those
> genes?
>
>
> Please, let me know, I highly appreciate your help in
> advance.
>
> Thank you,
> Saurin
>
> =====
> Saurin's WebWorld: http://hawkmail.monmouth.edu/~s0451884