[BioC] self-self hybridisations to estimate dye-bias

Nathan S. Watson-Haigh nathan.watson-haigh at csiro.au
Mon Mar 16 05:43:59 CET 2009

Hi all,

I'm new to 2-colour arrays and I've been reading with interest a few papers
about dye-bias:

Dobbin, K., Shih, J.H. & Simon, R., 2003. Statistical design of reverse dye
microarrays. Bioinformatics, 19(7), 803-810.

Dombkowski, A.A. et al., 2004. Gene-specific dye bias in microarray reference
designs. FEBS Letters, 560(1-3), 120-124.

Kerr, M.K., 2003. Design Considerations for Efficient and Effective Microarray
Studies. Biometrics, 59(4), 822-828.

Martin-Magniette, M. et al., 2005. Evaluation of the gene-specific dye bias in
cDNA microarray experiments. Bioinformatics, 21(9), 1995-2000.

I've also been thinking about self-self hybridisations (the same sample
hybridised to both Cy3 and Cy5):

Fang, H. et al., 2007. Self-self Hybridization As An Alternative Experiment
Design to Dye Swap for Two-color Microarrays. OMICS: A Journal of Integrative
Biology, 11(1), 14.

In the absence of any gene-dye interaction, the Cy3/Cy5 ratio should be zero in
the absence of gene*dye interaction.

I'm designing how best to hybridise our samples to Agilent chips. Our experiment
is simply a comparison between two groups with 5 biological replicates per group
(all 10 samples are biologically independent). From my reading, a multi-dye-swap
and loop design both have the same precision to detect differences between the 2
groups using the same number of arrays. However, both are 4 time more precise
(at best) than a reference design as relative abundances are measured directly,
rather than indirectly. I'm opting for the multi-dye-swap where group 1 rep 1 is
dye-swapped with group 2 rep 1, group 1 rep 2 is dye-swapped with group 2 rep 2
etc. This is because it is more robust in terms of missing data that may occur.
Am I also right in thinking that it could be extended more easily than a loop
design in the future - i.e. add additional biological replicates at a future date?

The idea behind the dye-swap is to get a measure of the error variance due to
dye-bias. I was wondering if were possible to obtain a better estimate by using
self-self hybridisations and if so, how to code this into limma.

For instance, my target file might look like:

FileName  Cy3  Cy5
1         G1   G2
2         G2   G1
3         G1   G2
4         G2   G1
5         G1   G2
6         G2   G1
7         G1   G2
8         G2   G1
9         G1   G2
10        G2   G1

Where G1 are biological replicates for Group 1 and G2 are biological replicates
for Group 2. The same two RNA samples are dye-swapped e.g. array 1 and 2 are
dye-swaps of the same RNA. Could I add, or even replace the dye-swaps, with
self-self hybs to provide a better estimate the dye-bias like this:

FileName  Cy3  Cy5
1         G1   G2
2         G2   G1
3         G1   G2
4         G2   G1
5         G1   G2
6         G2   G1
7         G1   G2
8         G2   G1
9         G1   G2
10        G2   G1
11        G1   G1
12        G2   G2

I'd appreciate thoughts and comments if you have them!


Dr. Nathan S. Watson-Haigh
OCE Post Doctoral Fellow
CSIRO Livestock Industries
Queensland Bioscience Precinct
St Lucia, QLD 4067

Tel: +61 (0)7 3214 2922
Fax: +61 (0)7 3214 2900
Web: http://www.csiro.au/people/Nathan.Watson-Haigh.html

More information about the Bioconductor mailing list