[BioC] Nimblegen arrays/Limma package:duplicate correlation and other problems

[r.athanasiadou]

Yes it is a tiling array with no duplicate spots.
I have looked into the packages that deal with tiling arrays like
"chIPchip"(is it available for windows yet?) but from what I could gather
from the way such packages work, they rely on having random fractionation of
the genome ie sonication. Unfortunately, my experiment required restriction
endonuclease digestion to fractionate the genome (produces specific and
predictable short fragments and ideally -no sequence bias- I expect a sharp
rise and fall of a positive region rather than a normal distribution of the
M-values) and I don't thing that the common algorithms to summarize the
probe-data are applicable in my case. 

I am thinking to rely on how many probes (out of the total number of probes
that should hybridize to each generated genomic fragment) give reproducible
and comparable results to summarize my probe-level data.

Niki



-----Original Message-----
From: Jenny Drnevich [mailto:drnevich at uiuc.edu] 
Sent: 27 March 2007 17:23
To: Sean Davis; bioconductor at stat.math.ethz.ch; r.athanasiadou
Subject: Re: [BioC] Nimblegen arrays/Limma package:duplicate correlation and
other problems

To add to Sean's keen observation...

Are these expression arrays or tiling arrays? I've never worked with 
Nimblegen arrays, but ~390,000 spots seems like a lot for expression 
arrays. Also, you said this is chIP-on-chip data, right? If they are 
expression arrays with multiple spots, then you will need to use 
duplicateCorrelation to estimate the spot-replicate correlations. If 
they are tiling arrays (usually used with chIP-on-chip data), you 
will need to analyze it a completely different way because probes 
that are close to each other on the chromosome will not have 
independent fluorescence values.

Jenny

At 11:00 AM 3/27/2007, Sean Davis wrote:
>On Tuesday 27 March 2007 11:13, Jenny Drnevich wrote:
> > Hi Niki,
> >
> > I guess it's not exactly clear in the help page for lmFit, but
> > correlation is only used if ndups > 1 or block is not NULL. Since you
> > shouldn't be using either of these, correlation won't be used and
> > hence the default value doesn't matter. This is all explained better
> > in the vignette, under Technical Replication.
>
>And just to be *absolutely* clear, you do not have replication on the
>array--is that correct?  It looked from the "genes" slot that there might
be
>replication, but it wasn't possible to tell.  If not, is there a "tiling"
>component to the array, such as tiling of the promoter regions?
>
>Sean

Jenny Drnevich, Ph.D.

Functional Genomics Bioinformatics Specialist
W.M. Keck Center for Comparative and Functional Genomics
Roy J. Carver Biotechnology Center
University of Illinois, Urbana-Champaign

330 ERML
1201 W. Gregory Dr.
Urbana, IL 61801
USA

ph: 217-244-7355
fax: 217-265-5066
e-mail: drnevich at uiuc.edu