[BioC] B value cutoff in LIMMA

[Gordon Smyth]

>Date: Fri, 7 Apr 2006 15:22:45 -0400
>From: Richard Friedman <friedman at cancercenter.columbia.edu>
>Subject: [BioC] B value cutoff in LIMMA
>To: "bioconductor at stat.math.ethz.ch List'"
>         <bioconductor at stat.math.ethz.ch>
>
>Dear Bioconductor List,
>
>         In an earlier posting, Gordon Smyth recommended
>a tentative cutoff of B>=0 for Limma output.
>Is it okay to publish Venn diagrams based on this
>cutoff or should the cutoff be validated by spot-checking with PCR?
>Is it okay to publish pathway overrepresentation
>analyses with this cutoff, or is PCR validation of the cutoff
>advised?
>
>Thanks and best wishes,
>Rich
>------------------------------------------------------------
>Richard A. Friedman, PhD
>Associate Research Scientist
>Herbert Irving Comprehensive Cancer Center
>Oncoinformatics Core
>Lecturer
>Department of Biomedical Informatics
>Box 95, Room 130BB or P&S 1-420C
>Columbia University Medical Center
>630 W. 168th St.
>New York, NY 10032
>(212)305-6901 (5-6901) (voice)
>friedman at cancercenter.columbia.edu
>http://cancercenter.columbia.edu/~friedman/
>
>
>"Cartesian duelism is when somebody told Decartes that he didn't think
>therefore he was, and Descartes challenged him to a duel".
>-Isaac Friedman, age 16

Well, this depends on journal policy and isn't for us to say. But, 
generally speaking, genomic analyses don't require validation while 
conclusions about individual genes do.

Genomic analyses are those which analyse the pattern of differential 
expression for all the genes on the array, e.g., Venn diagrams, 
cluster analyses, gene set enrichment tests, network analyses. It 
would be unreasonable to expect to validate every gene involved, and 
this should not  be scientifically necessary because the results 
should be robust to changes in a small number of genes.

On the other hand, if you are involved in gene discovery, and your 
conclusions concentrate on one or a few important genes, then good 
journals will almost certainly require validation.

There are probably a range of opinions on this.

BTW, there's nothing sacrosant about the B>0 cutoff, although this 
sort of level often gives a reasonable balance of false positives and 
false negatives.

Best wishes
Gordon