[BioC] 4-color microarrays

Henrik Bengtsson hb at maths.lth.se
Tue Sep 13 19:23:57 CEST 2005

A.J. Rossini wrote:
> On 9/9/05, *Henrik Bengtsson* <hb at maths.lth.se <mailto:hb at maths.lth.se>> 
> wrote:
>     A.J. Rossini wrote:
>      > Interesting -- does anyone know if you need to "compensate" for
>     4-color? (I
>      > know that you _sometimes_ have to for 4-color flow cytometry)
>      >
>      > best,
>      > -tony
>     Most likely, yes!  Use affine normalization to do it.  It works for two
>     or more channels.  I've got all methods implemented in aroma (search
>     google), but I'm about to release a lightweight version of this called
>     aroma.light.  For the moment see the below tech report (another version
>     submitted) at http://www.maths.lth.se/bioinformatics/publications/:
>     H. Bengtsson and O. Hössjer, Methodological study of affine
>     transformations of gene expression data with proposed normalization
>     method, Preprints in Mathematical Sciences 2003:38, Mathematical
>     Statistics, Lund University, 2003.
>     Quantile normalization a la RMA (Affy) would also do, but has more
>     degrees of freedom compared to the 2*nbrOfChannels-1 parameters for
>     affine normalization.
>     Henrik
>     PS. I'm going to MGED in Norway soon, so I'll be online less often. DS.
> Well, what I was suggesting was based on selection of dyes.  With 
> 4-color flow cytometry data, if the colors are not well selected, you 
> have to worry about "bleed-over", so bad (or "forced by manufacturer") 
> selections require that you may have to compensate (down-grade) for 
> bleed over from the other channels if the wavelengths of the colors are 
> too close or the selectivity is too wide.
> It's definitely true for 8-color and higher flow data, borderline for 
> 4/6 color.

Sorry not answering to your idea there; I was simply focusing on the 
"normalization" part.  I definitely agree that crosstalk becomes an 
important issue the more dyes you have within a giving wavelength 
region.  Even in Cy3/Cy5 data you do see a bit of crosstalk in the 
direction from the higher energy dye to the lower energy (emitted 
photons from one is absorbed by the other).  You could use similar 
methods as for cytometry data to correct for this type of crosstalk. 
However, in gene expression data you have to add some external controls 
in order to estimate the correction factor(s).  To just comment on the 
affine normalization: Doing an affine normalization (transformation) in 
(R,G) will not destroy your chances to correct for the crosstalk 
afterwards, but you could possibly incorporate in in one single 
estimation/correction step.

It would be interesting the see some multi-channel data, if anyone has 
it available.



> -- 
> best,
> -tony
> blindglobe at gmail.com <mailto:blindglobe at gmail.com>
> Muttenz, Switzerland.
> "Commit early,commit often, and commit in a repository from which we can 
> easily
> roll-back your mistakes" (AJR, 4Jan05).

More information about the Bioconductor mailing list