[BioC] dye swap vs. control channel
rdiaz at cnio.es
Fri Oct 24 11:05:14 MEST 2003
> I think it boils down to whether you are interested in a direct
> head-to-head comparison (then dye swapping is essential), or in more
> general, less direct statements about groups of samples, with the
> possibility of generalizing past the experiment (then choice of
> reference sample is critical).
Along these linese, if the researcher is certain that she is interested in
comparing two (and just these two) groups, then I prefer a dye-swap design.
If the question is not as well specified, if more groups could be of interest
in the future ---recall connected design issues---, or if they want to try to
find predictors (say, discrminant analysis) or, (in spite of admonitions
against it) use clustering, I prefer to use a standard (a pool --below),
always in the same channel. I think with this approach we tend to view the
standard or reference as something we need to live with with cDNA arrays, not
something we really care about or wish we had to use ---I guess we wish we
had affy arrays.
> in, given the associated costs. Right now, if you have a good
> reference or reference prototype for your situation, and can afford
> biological replications, I'd probably do reference comparisons.
When a reference is used, and seen as a necessary experimental evil, I often
want it to be as representative as possible, but at the same time as little
variable as possible (i.e., for the reference use aliquots from the pool so
that variation from pool to pool sample is nearly inexistent). This is
something we have done: form the pool using samples from, say, 30 normal
subjects; then, "validate" the pool by using another set of normal samples
(not used in the construction of the pool); if the pool is OK, then we should
see no genes differentially expressed. (The size of the set of normal samples
for "validation" is choosen so that it is at least a little bit larger than
the sample size of the experimental conditions).
> So all of a sudden, we've moved beyond statistical design, to
> cost-function based decision theory and risk analysis.
> Sorry, no easy answers this morning.
> I'd be interested in hearing others opinions on the topic as well.
> Isaac Mehl <imehl at ucsd.edu> writes:
> > this design brings up a question i always have. is it "better" to do
> > all experiments in one channel (Cy5) and compare every sample to a
> > standard (cy3)? this way you can use less arrays or do more biological
> > replicates. IMHO getting repeated measurements of biological variation
> > is more important than dye swap.
> > very interested to hear what people think about this topic since it is
> > integral to experimental design.
> > -isaac
> > DIF1,2 and 3 are different but similar drugs...............
> >>> Slides 1-6 are treatment 1 (DIF1) Vs No treatment
> >>> Slide1 Cy5/Cy3 (DIF1/no treatment)
> >>> Slide2 Cy3/Cy5 (DIF1/no treatment)
> >>> Slide3 Cy5/Cy3 (DIF1/no treatment)
> >>> Slide4 Cy3/Cy5 (DIF1/no treatment)
> >>> Slide5 Cy3/Cy5 (DIF1/no treatment)
> >>> Slide6 Cy5/Cy3 (DIF1/no treatment)
> >>> Slides 7-12 are treatment 2 (DIF2) Vs No treatment
> >>> Slide7 Cy5/Cy3 (DIF2/no treatment)
> >>> Slide8 Cy3/Cy5 (DIF2/no treatment)
> >>> Slide9 Cy5/Cy3 (DIF2/no treatment)
> >>> Slide10 Cy3/Cy5 (DIF2/no treatment)
> >>> Slide11 Cy3/Cy5 (DIF2/no treatment)
> >>> Slide12 Cy5/Cy3 (DIF2/no treatment)
> >>> Slides 13-18 are treatment 3(DIF3) Vs No treatment
> >>> Slide13 Cy5/Cy3 (DIF3/no treatment)
> >>> Slide14 Cy3/Cy5 (DIF3/no treatment)
> >>> Slide15 Cy5/Cy3 (DIF3/no treatment)
> >>> Slide16 Cy3/Cy5 (DIF3/no treatment)
> >>> Slide17 Cy3/Cy5 (DIF3/no treatment)
> >>> Slide18 Cy5/Cy3 (DIF3/no treatment)
> >>> I'd obviously like to compare across the different treatments DIF1,2
> >>> and 3
> > --
> > -isaac mehl
> > gene expression lab (gele)
> > salk institute
> > 10010 n. torrey pines rd.
> > la jolla ca. 92037
> > http://genex.salk.edu
> > _______________________________________________
> > Bioconductor mailing list
> > Bioconductor at stat.math.ethz.ch
> > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
Centro Nacional de Investigaciones Oncológicas (CNIO)
(Spanish National Cancer Center)
Melchor Fernández Almagro, 3
28029 Madrid (Spain)
PGP KeyID: 0xE89B3462
More information about the Bioconductor