[BioC] Agilent spike-in probes

Sean Davis sdavis2 at mail.nih.gov
Fri Apr 4 13:46:06 CEST 2008 

On Fri, Apr 4, 2008 at 7:20 AM, Srinivas Iyyer
<srini_iyyer_bio at yahoo.com> wrote:
> 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.

MAList and RGList objects subset just like other matrices and
data.frames.  Assuming that you know what probes are control probes,
you can simply create a new MAList, removing the control probes.

MA2 <- MA[-controlProbes,]

where controlProbes is a vector of the indices of control probes.

Sean


>
>  --- 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]]
>
>  _______________________________________________
>  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
>

More information about the Bioconductor mailing list