Hello Dr. Tamir
I'm working with nimblegen tiling arrays and I am dealing with some
normalization issues. In looking into the problem, I noticed a thread
on the bioconductor message board where you stated that after some
tests and discussion, you advise against loess normalization when
using these arrays. I too have noticed some problems (via the
MAPlots) when I used Loess normalization. Qualitatively I'd say that
my data looks better and more reproducible when I only use the
Nimblegen normalization. Could you elaborate on the issues that
you've noticed when using loess on the nimblegen arrays??
Best Regards
Simon Knott
PHD Student
University of Southern California
Department of Computational and Molecular Biology
On Wed, Mar 19, 2008 at 11:16 AM, Simon Knott <knott at="" usc.edu="">
wrote:
> Hello Dr. Tamir
>
> I'm working with nimblegen tiling arrays and I am dealing with some
> normalization issues. In looking into the problem, I noticed a
thread
> on the bioconductor message board where you stated that after some
> tests and discussion, you advise against loess normalization when
> using these arrays. I too have noticed some problems (via the
> MAPlots) when I used Loess normalization. Qualitatively I'd say
that
> my data looks better and more reproducible when I only use the
> Nimblegen normalization. Could you elaborate on the issues that
> you've noticed when using loess on the nimblegen arrays??
There is no problem with using loess on Nimblegen arrays. However,
there is a problem with using loess on ChIP-chip experiments. There
is an expected nonlinearity in a good ChIP-chip experiment with more
high-ratio than low-ratio spots in the well-measured spectrum of the
MA plot. In other words, there may be a bowing up of the MA plot that
is expected. If one uses loess normalization, it will effectively
remove the bowing and make the MA plot "look better", but you will
have removed your signal. We typically do not really attempt to
normalize further ChIP-chip experiments. That said, what
normalization you choose (or not) will be somewhat dependent on what
analysis method you will be using to find peaks.
Sean