[BioC] Evidence-based guidelines for use the method="robust" option in lmFit

[Richard Friedman]

Dear list,

	I would greatly appreciate  guidance on when to use the  
method="robust" option in lmFit.
Use or non-use of the methods produces vastly different results on the  
set which I am studying.
Below are reproduced 2 notes, from the list. The  first from Gordon  
Smyth stating that there is no
rule when to use it, and the second from Jenny Drenivch suggesting  
that sample size be a guide..
I would like to then ask the question a little differently: Has the  
effectiveness of using, method="robust"
been evaluated in any study? The original Limma paper showed the  
superiority of Limma
over the unmoderated t-statistic for ranking genes. Subsequent studies  
showed how Limma is superior
to the t-test iusing simulation data. Has any paper compared the  
relative merits of robust and standard fitting
in Limma? Or alternatively, has anyone ever obtained results with  
robust fitting that were validated
independently but were not found by ordinary fitting?

Here are the 2 previous posts:

##########################################

Gordon Smyth smyth at wehi.edu.au
Fri Apr 8 13:33:28 CEST 2005

     * Previous message: [BioC] no BioConductor posting guide
     * Next messag

Some people always like to use robust statistical methods when analysing
microarray data, and the "robust" option to lmFit() is provided for this
reason. There is no rule which tells you which data to use it for and  
which
not.

Gordon


Jenny Drnevich drnevich at illinois.edu
Thu Jan 8 18:58:21 CET 2009

     * Previous message: [BioC] lmFit function
     * Next message: [BioC] lmFit function
     * Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Priscila,

The robustspline method for normalization has
nothing to do with the lmFit(method="robust").
lmFit can either fit the model using a least
squares regression or a robust regression, which
down-weights replicates that are different from
the other replicates. Whether or not to use
lmFit(method="robust") doesn't depend on which
normalization method you use, but rather (IMO)
how many replicates you have. If you have a
relatively large number of replicates, say 6 or
more, then the robust fitting of the model may
help to remove true outliers from affecting the
data. However, if you only have 3 replicates, as
is usual for microarray experiments, using the
robust estimation may remove real variation in
your samples and lead to more false-positives.

That's my take on the situation...
Jenny


##########################################

Thanks and best wishes,
Rich
------------------------------------------------------------
Richard A. Friedman, PhD
Associate Research Scientist,
Biomedical Informatics Shared Resource
Herbert Irving Comprehensive Cancer Center (HICCC)
Lecturer,
Department of Biomedical Informatics (DBMI)
Educational Coordinator,
Center for Computational Biology and Bioinformatics (C2B2)/
National Center for Multiscale Analysis of Genomic Networks (MAGNet)
Room 824
Irving Cancer Research Center
Columbia University
1130 St. Nicholas Ave
New York, NY 10032
(212)851-4765 (voice)
friedman at cancercenter.columbia.edu
http://cancercenter.columbia.edu/~friedman/

I am a Bayesian. When I see a multiple-choice question on a test and I  
don't
know the answer I say "eeney-meaney-miney-moe".

Rose Friedman, Age 14