[BioC] SAM and qvalue

[Holger Schwender]

Hi Nicolas,

these differences are due to the differing calculation of the FDR and the q-value. The FDR is computed using the observed and expected d values that fall outside the interval (cutlow, cutup), whereas the q-values are computed as in the R package qvalue and based on the SAM p-value which uses symmetric thresholds, i.e., e.g., (-cutup, cutup). So it can and will happen that not all q-value estimates are smaller than the FDR value if, e.g., |cutup|>|cutlow|.

Best,
Holger

-------- Original-Nachricht --------
Datum: Mon, 06 Nov 2006 16:38:23 +0100
Von: Nicolas Servant <Nicolas.Servant at curie.fr>
An: Bioconductor <bioconductor at stat.math.ethz.ch>
Betreff: [BioC] SAM and qvalue

> Hi all,
> 
> I have a question about SAM (siggenes) and its adjusted pvalues (qvalues).
> When i perform SAM on Golub data for a FDR threshold = 5%:
> 
> FDR=5/100
> output<-sam(golub,golub.cl,rand=123,delta=seq(0.1,5,0.05))
> res.sum<-summary(output)
> sum.output <- res.sum at mat.fdr
> delta<-sum.output[sum.output[,"FDR"]<=FDR,][1,"Delta"]
> delta.sum<-summary(output,delta)
> ds.mat.sig <- delta.sum at mat.sig
> 
> I found 894 significant genes.
>     Row   d.value      stdev      p.value      q.value    R.fold
> 1    829  8.165222 0.29582512 0.000000e+00 0.000000e+00 7.2771792
> 2   2124  7.964784 0.17786969 0.000000e+00 0.000000e+00 3.3953035
> 3   2600  6.102371 0.19112194 0.000000e+00 0.000000e+00 2.6686992
> ....
> 892  142 -1.689638 0.11912464 3.305801e-02 5.393673e-02 0.8178807
> 893  864 -1.689047 0.08528524 3.312029e-02 5.393673e-02 0.8312349
> 894  686 -1.689045 0.20350807 3.312029e-02 5.393673e-02 0.7272737
> 
> For a FDR threshold, SAM use the Delta, the cutlow and the cutup values 
> to find significant genes.
> How can we explain that the last genes of my list have a qvalue bigger 
> than 5% (my FDR threshold) ?
> I notice that their dstatistics are in the good range (cutlow-cutup), It 
> certainly explains why these genes are significants.
> 
> Thanks for your help !
> Best Regards,
> 
> Nicolas
> 
> -- 
> Nicolas Servant
> Equipe Bioinformatique
> Institut Curie 
> 26, rue d'Ulm - 75248 Paris Cedex 05 - FRANCE
> 
> Email: Nicolas.Servant at curie.fr
> Tel: 01 53 10 70 55
> http://bioinfo.curie.fr/
> 
> _______________________________________________
> 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

--