[BioC] Analysis of Affymetrix Human Gene 2.0 ST arrays
James W. MacDonald
jmacdon at uw.edu
Fri Nov 29 20:33:36 CET 2013
Hi Maria,
Please don't take messages off-list (e.g., use Reply-all). We like to
think of the list archives as a repository of information that people
can search, and if messages become private, that hampers the usefulness
of the archives.
On 11/29/2013 1:38 PM, María Maqueda González wrote:
> Hi Jim,
> Many thanks for your quick and very comprehensive response.
>
> From your comments, I have one more question related:
>
> (1) I understand your comments about the intron control transcripts,
> but I do not fully understand the rescue transcript category that I
> have also obtained in my topTable transcripts.
There are two things to think about here.
First, there is the issue of statistical significance versus biological
significance. Note that the t-statistic is a fraction, and in the
numerator you have the difference between the means of two groups, and
in the denominator you have the standard error of that difference. The
standard error is based on the intra-group variability. So if you have a
particular probeset and the intra-group variability for that probeset is
extremely small, then you can end up with a statistically significant
result even if the fold change isn't very large at all.
The eBayes step is intended to protect against this to some extent, by
adjusting 'too small' standard errors towards the overall variance
estimate, but protecting against something and completely eliminating it
are two different things. So it may be that the differences for these
controls aren't that great, and it is just happenstance that the
intra-group variance is small enough to get statistical significance.
One thing you can do to protect against that sort of thing is to filter
out probesets that don't really change expression very much in any
samples (or just use getMainProbes and nuke all these controls in the
first place, which is what I would do).
Second, just because something shows up in a topTable, doesn't mean it
is actually differentially expressed. I don't know how you are adjusting
for multiple comparisons, but let's just assume you are using FDR. If
you then take the probesets with an FDR > 0.05, you are accepting that
up to 5% of the probesets in that list are false positives. In other
words, 5% of the probesets in that table aren't really differentially
expressed, they just happen to have a large t-statistic by chance. Thus,
the rescue probeset(s) that you have might just be false positives.
Best,
Jim
>
> No need to send the function, but thanks in any case for offering.
>
> Regards,
>
> Maria
>
> > Date: Fri, 29 Nov 2013 09:04:20 -0500
> > From: jmacdon at uw.edu
> > To: guest at bioconductor.org
> > CC: bioconductor at r-project.org; mmaqueda at live.com
> > Subject: Re: [BioC] Analysis of Affymetrix Human Gene 2.0 ST arrays
> >
> > Hi Maria,
> >
> >
> > On 11/29/2013 6:18 AM, María Maqueda [guest] wrote:
> > > Dear all,
> > >
> > > I am analyzing a set of Affymetrix Human Gene 2.0 ST arrays, this
> is my first time working with this type of arrays so I have a few
> general questions. I would very much appreciate any advice you could give.
> > >
> > > (1) I have obtained different lists of differentially expressed
> genes (using eBayes() from limma). In those lists, some control
> transcripts are popping up (i.e normgene -> intron category among
> other categories). I was not expecting this type of transcripts at
> this point. In theory after normalization, no control transcripts
> should appear, am I right? Have you experienced this?
> > > I have read that one possibility is to use getMainProbes before
> topTable selection but I wonder if there could be something wrong from
> the beginning with my normalization process (I have used rma() â
> transcript level - from oligo). What is your opinion?
> >
> > I don't think it has anything to do with the normalization. Instead, I
> > think it is a combination of poorly designed probes and highly
> expressed
> > genes for which there are sufficient unprocessed mRNA transcripts that
> > still have their introns intact (remember that the processing of
> samples
> > stops all enzymatic activity very quickly as a first step, so any mRNA
> > that is in the process of being transcribed, or is just finishing
> > transcription will likely still have introns).
> >
> > >
> > > (2) This type of arrays also includes lincRNA transcripts and I am
> interested in considering them for my analysis. The thing is that I am
> using hugene20sttranscriptcluster.db for annotation and these lincRNA
> are not included. Would this library be able to handle them?
> >
> > Hypothetically yes, as of now not really. It doesn't seem like that
> many
> > have been annotated with Entrez Gene IDs, and until that happens they
> > won't appear in the annotation packages. And even for those that do
> have
> > Entrez Gene IDs, the information stops there - you go to NCBI and it
> > just says that the lincRNA is supposed to exist, but nothing else.
> >
> > >
> > > (3) I tried to make my own annotation package thru makeDBPackage
> based on .csv annotation file from Affy but I got an errorâ¦: Error
> in `[.data.frame`(csvFile, , GenBank IDName) : undefined columns selected
> > > I have already read in this mailing list that makeDBPackage may
> expect a HGU133plus2 annotation âstyleâ. Would the library
> annotationForge be able to handle this?
> >
> > AnnotationForge cannot handle the csv files for these arrays directly,
> > as they are completely different from the old style 3'-biased arrays
> > like the hgu133plus2 that you mention. I have a function I can give you
> > to make the input file for the annotation package, but I don't think it
> > is worth it because it would be the function that I already used to
> make
> > the annotation package you can get from BioC. So you could go through
> > all the effort to make something you can already get.
> >
> > But if you want it, I will send it to you.
> >
> > Best,
> >
> > Jim
> >
> >
> > >
> > >
> > > Many thanks in advance for any help!
> > >
> > >
> > > MarÃa Maqueda
> > >
> > > Biomedical Engineering Research Centre (CREB)
> > > Universitat Politècnica de Catalunya (UPC)
> > >
> > > -- output of sessionInfo():
> > >
> > >> sessionInfo()
> > > R version 3.0.1 (2013-05-16)
> > > Platform: x86_64-w64-mingw32/x64 (64-bit)
> > >
> > > locale:
> > > [1] LC_COLLATE=Spanish_Spain.1252 LC_CTYPE=Spanish_Spain.1252
> > > [3] LC_MONETARY=Spanish_Spain.1252 LC_NUMERIC=C
> > > [5] LC_TIME=Spanish_Spain.1252
> > >
> > > attached base packages:
> > > [1] parallel stats graphics grDevices utils datasets methods base
> > >
> > > other attached packages:
> > > [1] human.db0_2.9.0 AnnotationForge_1.2.2
> > > [3] hugene20sttranscriptcluster.db_2.12.1 org.Hs.eg.db_2.9.0
> > > [5] AnnotationDbi_1.22.6 BiocInstaller_1.12.0
> > > [7] limma_3.16.8 pd.hugene.2.0.st_3.8.0
> > > [9] oligo_1.24.2 Biobase_2.20.1
> > > [11] oligoClasses_1.22.0 BiocGenerics_0.6.0
> > > [13] RSQLite_0.11.4 DBI_0.2-7
> > >
> > > loaded via a namespace (and not attached):
> > > [1] affxparser_1.32.3 affyio_1.28.0 annotate_1.38.0
> > > [4] Biostrings_2.28.0 bit_1.1-10 codetools_0.2-8
> > > [7] ff_2.2-12 foreach_1.4.1 genefilter_1.42.0
> > > [10] GenomicRanges_1.12.5 IRanges_1.18.4 iterators_1.0.6
> > > [13] preprocessCore_1.22.0 splines_3.0.1 stats4_3.0.1
> > > [16] survival_2.37-4 tools_3.0.1 XML_3.98-1.1
> > > [19] xtable_1.7-1 zlibbioc_1.6.0
> > >
> > > --
> > > Sent via the guest posting facility at bioconductor.org.
> > >
> > > _______________________________________________
> > > Bioconductor mailing list
> > > Bioconductor at r-project.org
> > > https://stat.ethz.ch/mailman/listinfo/bioconductor
> > > Search the archives:
> http://news.gmane.org/gmane.science.biology.informatics.conductor
> >
> > --
> > James W. MacDonald, M.S.
> > Biostatistician
> > University of Washington
> > Environmental and Occupational Health Sciences
> > 4225 Roosevelt Way NE, # 100
> > Seattle WA 98105-6099
> >
--
James W. MacDonald, M.S.
Biostatistician
University of Washington
Environmental and Occupational Health Sciences
4225 Roosevelt Way NE, # 100
Seattle WA 98105-6099
More information about the Bioconductor
mailing list