[BioC] analysis of 70 groups for differential expression
smyth at wehi.EDU.AU
Thu Sep 20 02:12:39 CEST 2007
>Date: Tue, 18 Sep 2007 10:50:22 -0400
>From: "Artur Veloso" <abveloso at gmail.com>
>Subject: Re: [BioC] analysis of 70 groups for differential expression
>To: "Bioconductor List" <bioconductor at stat.math.ethz.ch>
>I have a question very similar to the one described here.
>I have samples from 11 environmental groups, so there is no control in the
>experiment. Therefore, there are 55 comparisons that can be made and are of
>interest in the experiment.
>My approach before reading this was to run all comparisons independently
>from each other, using makeContrasts, contrasts.fit,etc 55 times, and then
>use decideTests to pull out the differently expressed genes. But I believe
>that this is incorrect, since I am not adjusting the p-value for these 55
>The new approach that I tried was to run all the comparisons together (use
>makeContrasts just once) and then use decideTests and use each column of the
>TestResult matrix to extract DE genes for that specific comparison.
>But, when I compared TestResult matrix columns from the same comparisons run
>under the different methods described above I found no differences.
See the help page ?decideTests, which tells you that "setting
method="separate" is equivalent to using topTable separately for each
coefficient in the linear model fit". In other words, the default
method for decideTests() is the same as testing each contrast
separately, as you have found.
If you want to adjust for multiple testing across the 55 contrasts,
you need to use decideTests() with another method, e.g., method="global".
> But I
>did not look into the q-values or the lods.
>Was I wrong to run the 55 comparisons separately? Is the new approach more
>Thank you very much for the help!
>Artur B. Veloso
>Masters in Marine Biology Candidate
>College of Charleston, South Carolina, USA
More information about the Bioconductor