[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