Hi,
has anybody experience with Chromatin IP on Chip normalizations?
We use an oligo tiling array as well as a ~500bp DNA fragments
spotted array (which is my priority currently).
Are there special normalization procedures in some of the packages?
I couldn't find any of those yet.
Thank you very much for your answers
Ido M. Tamir
Hi Ido,
> has anybody experience with Chromatin IP on Chip normalizations?
> We use an oligo tiling array as well as a ~500bp DNA fragments
> spotted array (which is my priority currently).
>
> Are there special normalization procedures in some of the packages?
> I couldn't find any of those yet.
Normalization always needs to be adapted to the particular platform,
quality issues etc. of the arrays at hand, but why do you think you
need
a special normalization procedure for ChIP compared to RNA
hybridization?
We have successfully used vsn (in the package of the same name) for
normalizing ChIP data on two-color spotted arrays, and I know other
people have used loess-normalization (marray or limma packages) as
well.
--
Best regards
Wolfgang
-------------------------------------
Wolfgang Huber
European Bioinformatics Institute
European Molecular Biology Laboratory
Cambridge CB10 1SD
England
Phone: +44 1223 494642
Fax: +44 1223 494486
Http: www.ebi.ac.uk/huber
On May 25, 2005, at 10:06 AM, Wolfgang Huber wrote:
>
> Hi Ido,
>
>> has anybody experience with Chromatin IP on Chip normalizations?
>> We use an oligo tiling array as well as a ~500bp DNA fragments
>> spotted array (which is my priority currently). Are there special
>> normalization procedures in some of the packages?
>> I couldn't find any of those yet.
>
> Normalization always needs to be adapted to the particular platform,
> quality issues etc. of the arrays at hand, but why do you think you
> need
> a special normalization procedure for ChIP compared to RNA
> hybridization?
In data we have looked at, the centrality paramater (whatever that is)
is often not correct, as chIPchip data is effectively one-sided.
This,
of course, depends on the experiment (high enrichment leads to a
skewed
distribuiton of ratios). In any case, normalization will affect any
downstream analyses you perform, so you need to understand how the
normalization you choose might impact your ability to see "positive"
probes.
Sean
Hi Sean,
> In data we have looked at, the centrality paramater (whatever that
is)
> is often not correct, as chIPchip data is effectively one-sided.
This,
> of course, depends on the experiment (high enrichment leads to a
skewed
> distribuiton of ratios).
Agree. It might be worthwhile in this context to look at the "midpoint
of the shorth" as a centrality parameter (function "shorth" in the
genefilter package), it is often less susceptible to asymetric tails
than mean or median (see also the example in the man page).
The outlier detection in "vsn" tries to be quite robust against such
asymmetric tails (and there is a simulation study in the paper cited
below about this). I think loess also is robust against these to a
certain extent, but not so sure how much.
[1] http://www.bepress.com/sagmb/vol2/iss1/art3/
Best regards
Wolfgang
-------------------------------------
Wolfgang Huber
European Bioinformatics Institute
European Molecular Biology Laboratory
Cambridge CB10 1SD
England
Phone: +44 1223 494642
Fax: +44 1223 494486
Http: www.ebi.ac.uk/huber
