[BioC] Agilent spike-in probes
Srinivas Iyyer
srini_iyyer_bio at yahoo.com
Fri Apr 4 13:20:39 CEST 2008
Dear Naomi,
How can I remove control probes before doing
expression analysis.
i am following some steps lile the followings.
RG <- read.maimages(files$targes,'agilent')
RG.b <-
backgroundSubtraction(RG,method='normexp',offset=50)
MA <- normalizewithinarrays(RG.b,method='loess').
Where in these steps I can remove controls entirely.
Thank you.
Srini
--- Naomi Altman <naomi at stat.psu.edu> wrote:
> In my experience with Agilent arabdopsis arrays,
> some of the Agilent
> spike-ins bind only to one of the dyes (or bind much
> more strongly to
> one). I always remove the controls before doing
> differential
> expression analysis.
>
> Naomi
>
>
> At 08:29 AM 3/30/2008, Sean Davis wrote:
> >On Sat, Mar 29, 2008 at 11:39 PM, Srinivas Iyyer
> ><srini_iyyer_bio at yahoo.com> wrote:
> > > dear sean,
> > > i apologize for sending this email and attached
> > > figures to you. I am not sure if I can send
> figures as
> > > attachment to mailing list. I wanted to see
> expert
> > > opinion on this particular topic because this
> is first
> > > time i am analyzing agilent chip data.
> > > Would you please look into my design, code and
> figures
> > > and let me know if this method okay.
> > >
> > > Spike-in probes are for QC purposes, if so why
> I am
> > > getting spike-in probes as top candidates. Is
> there a
> > > way to suppress them.
> > > Thank you and I appreciate your help.
> > >
> > >
> > >
> > > dear group,
> > >
> > > I have agilent 4x44 (G4112F) chips. the hybs
> are done
> > > as a paired design. sample obtained from
> patient
> > > before and after treatment. 40 patient are in
> the
> > > study. chip was hybridized with before
> treated(cy3)
> > > and after treated (cy5) rna.
> > >
> > > I used LIMMA for normalizing and to calcuate
> > > differentially expressed.
> > >
> > > in the first step, I did not go for background
> > > subtraction and observed a blown-out ma plot.
> >
> >I'm not sure what "blown-out" means, but Agilent
> typically does
> >background subtraction automatically (you'll need
> to look at the
> >specific image extraction protocol to check). If
> you use the
> >gProcessedSignal and rProcessedSignal (these are
> not the defaults in
> >limma), you will probably get the benefit of their
> spatially-detrended
> >loess background subtraction.
> >
> > > when i did background subtraction, i observed a
> more
> > > compact ma. For q-q plot points at intersection
> are
> > > not many suggesting that many genes are
> differentially
> > > expressed. (figures are attach
> > >
> > > my main concern is, of top100 (from toptable
> > > number=100), most of the probesets are spikein
> > > probesets. (+)E1A_r60_a22 , DCP_22_6,DCP_22_7
> and so
> > > on.
> >
> >This could be dye bias, but I'm not sure. You
> didn't do dye swaps, so
> >you cannot separate signal from dye bias. In any
> case, you will need
> >to do some QC. Agilent provides a huge amount of
> QC and plots on the
> >scanner machine. You can always look there to see
> what they do.
> >Also, their technical manuals are pretty good at
> giving direction
> >about the technology and the array data processing.
> >
> > > These spike-in probes are highly differentlly
> > > expressed.
> > >
> > >
> > > my targets file
> > >
> > > filename cy3 cy5
> > > patient1 before after
> > > patient2 before after
> > > ......
> > > patient40 before after
> > >
> > > my design matrix:
> > > desin <- modelMatrix(targets,ref='before')
> > > > desin
> > > after
> > > [1,] 1
> > > [2,] 1
> > > [3,] 1
> > > [4,] 1
> > > [5,] 1
> > > [6,] 1
> > > [7,] 1
> > > [8,] 1
> > >
> > > RG2 <- backgroundCorrect(RG,method='subtract')
> > > MA2 <-
> normalizeWithinArrays(RG2,method='loess')
> > > plotDensities(MA2)
> > > boxplot(MA2$M~col(MA2$M),names=colnames(MA2$M))
> > > MA2a <-
> normalizeBetweenArrays(MA2,method='scale')
> >
> >These are two-color arrays. Do you really need to
> do the
> >between-array normalization? You might, but I
> think you might spend
> >some time proving to yourself that is the case.
> >
> > > fit.b <- lmFit(MA2a,design)
> > > fit.b <- eBayes(fit.b)
> > >
>
topTable(fit.b,number=50,adjust.method='BH')[,c(5,9,10,11,12,13)]
> > >
> > > my questions are:
> > >
> > > 1. for this paired sample (cy3,cy5) design, is
> my
> > > limma model matrix okay.
> > > 2. how to avoid getting spike-in . I never saw
> > > spike-in getting into top-table. is there some
> mistake
> > > going on at some place. is it normal for
> spike-in
> > > probes to come as top differentially expressed
> probes.
> >
> >It happens, yes. I would definitely do some QC,
> though. It doesn't
> >look like you have done any in your code here.
> >
> > > 3. are the attached figures (MA plot and q-q
> plot)
> > > reflect a good normalized data.
> >
> >The qq plot does not really tell you about
> normalization. The single
> >MA plot looks OK. You will want to look at all of
> the MA plots and
> >some more extensive QC.
> >
> > > 4. my chip is hgug4112F. I do not see
> annotation file
> > > on bioconductor.
> >
> >I think the hgug4112a annotation package is what
> you want. You'll
> >want to double-check that with a few lookups to be
> sure.
> >
> >Sean
> >
> >_______________________________________________
> >Bioconductor mailing list
> >Bioconductor at stat.math.ethz.ch
> >https://stat.ethz.ch/mailman/listinfo/bioconductor
> >Search the archives:
>
>http://news.gmane.org/gmane.science.biology.informatics.conductor
>
> Naomi S. Altman
> 814-865-3791 (voice)
> Associate Professor
> Dept. of Statistics
> 814-863-7114 (fax)
> Penn State University
> 814-865-1348 (Statistics)
> University Park, PA 16802-2111
>
>
____________________________________________________________________________________
[[elided Yahoo spam]]
More information about the Bioconductor
mailing list