[BioC] Large # of significant genes with SAM

[Wolfgang Huber]

Hi Vincent,

after all the good answers, here some more comments:

In one of our papers that compared 37 matched normals and tumors, we 
also found large numbers. Have a look at Fig. 3A of PubMed-ID 11691851, 
which shows that in this experiment the number of "significantly 
differentially expressed genes" growed linearly (!) with the number of 
samples, for up to 37. At the time, we were similarly surprised.

Basically, the reason is that t-test (on which SAM is based) looks for 
differences in the mean between tumor and normal - however small, as 
long as it significant.

It is important to distinguish "effect size" from "significance".

There is an excellent paper on this subject: Pepe MS, Longton G, 
Anderson GL, Schummer M. Selecting differentially expressed genes from 
microarray experiments. Biometrics. 2003 Mar;59(1):133-42.
PMID: 12762450

Their pAUC statistics is implemented in the ROC package (but slow...)

Also have a look at the exercise "Testing for Differential Expression" 
(Wed morning) of our 2004 bioC short course:
http://www.bioconductor.org/workshops/Bressanone

Best wishes
   Wolfgang


> I am finding using siggenes' SAM @ q<0.05 (26 samples on cDNA chips)
> that 37% of all genes are regulated with respect to patient-matched
> "normal" tissues in somme tumors not particularly known for huge
> aneuploidy. Looking at another data set from the same cancer but
> collected by another group on indepentent samples on Affy, I got 34%.
> The number seems to hold.
> 
> How to interpret this? Are really 30% of the genes disturbed, even to
> a small extent, in these tumors? Could SAM do something wrong? If yes,
> how to verify it?
> 
> Any advise, shared experience, references, etc. are welcome
> 
> Cheers
> 
> Vincent
> 
> 
> ------------------------------------------
> Vincent Detours, Ph.D.
> IRIBHM
> Bldg C, room C.4.116
> ULB, Campus Erasme, CP602
> 808 route de Lennik
> B-1070 Brussels
> Belgium
> 
> Phone: +32-2-555 4220
> Fax: +32-2-555 4655
> 
> E-mail: vdetours at ulb.ac.be
> 
> URL: http://homepages.ulb.ac.be/~vdetours/
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor


-- 
Best regards
   Wolfgang

-------------------------------------
Wolfgang Huber
European Bioinformatics Institute
European Molecular Biology Laboratory
Cambridge CB10 1SD
England
Phone: +44 1223 494642
Fax:   +44 1223 494486
Http:  www.ebi.ac.uk/huber