[BioC] SAM siggenes number of permutations

[Holger Schwender]

Hi Olivier,

for Delta=2.5 you do not have 9 false positives, you have -- what Tusher et al. call -- 9 falsely called genes. The expected number of false positives is given by pi0 * False. So you have an expected number of false positives of 0.78 * 9. I know that this is pretty confusing and I will add a more detailed description on this to the siggenes vignette.
Moreover, the FDR is computed by pi0 * False / Called which is here 0.0209. So your estimated FDR is 2.1%.

Best,
Holger

-------- Original-Nachricht --------
> Datum: Wed, 30 Jan 2008 19:03:47 +0100
> Von: "olivier armant" <olivier.armant at itg.fzk.de>
> An: bioconductor at stat.math.ethz.ch
> Betreff: Re: [BioC] SAM siggenes number of permutations

> Dear all,
> 
> Thank you for your feedback!
> Everyone can guess that I am not a statistician....
> :)
> 
> Here is the results I get with my samples running SAM.
> > sam.out<-sam(normalized, sam.c1, B=100,var.equal=TRUE, med=TRUE)
> 
> SAM Analysis for the Two-Class Unpaired Case Assuming Equal Variances 
>  
> Delta   p0 False Called    FDR
> 1    0.1 0.78 16417  26195 0.4887
> 2    1.3 0.78    78    751 0.0810
> 3    2.5 0.78     9    335 0.0209
> 4    3.7 0.78     3    146 0.0160
> 5    4.9 0.78     1     71 0.0110
> 6    6.1 0.78     0     21      0
> 7    7.3 0.78     0      6      0
> 8    8.5 0.78     0      4      0
> 9    9.8 0.78     0      3      0
> 10  11.0 0.78     0      0      0
> 
> For me having 335 genes with only 9 false positve is already satisfactory.
> Even 20% FDR would be ok as I will have to confirm the microarray data by
> other methods...
> Do you think I can use those data as a first way to select candidate
> genes? I would say yes, but I would like advise from the specialists to see if I
> am not doing something wrong!!
> Is it better to use the limma package that you suggested??
> 
> Cheers
> 
> Olivier
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives:
> http://news.gmane.org/gmane.science.biology.informatics.conductor

--