[BioC] Agilent spike-in probes

[Sean Davis]

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