[BioC] Three-color microarray analysis

Wolfgang Huber whuber at embl.de
Mon Oct 18 06:07:42 CEST 2010

Hi Tom Sing,

Thank you, this sounds reasonable. I think the analysis is conceptually 
not different from that of 4 replicates of 3 one-color arrays.

So, you could think of the measurement for each gene as a point in 3D 
space, and consider different projections (e.g. on the plane normal to 
the average vector (1,1,1)), perhaps like Fig. 7 in 

Depending on how that plot looks, one could attempt to detect gene set 
enrichment in different areas (directions) of the plot, e.g. using 
Hotelling's t-statistic and the applyByCategory function in the Category 
package; or the polar angle.

Hope this helps.


PS - was a dye swap performed between the four biological replicates? If 
not, I'd expect to pay a substantial amount of attention to confounding 
of biological effects with dye-effects.

Il Oct/17/10 7:02 PM, Thomas Sandmann ha scritto:
> Dear Wolfgang,
> thanks a lot for your pointers to all the different packages.
> To give you a bit more information about the experiment:
> In this study, two factors are investigated: genotype and food.
> Three different treatments were performed:
> A) wt genotype + normal food
> B) wt genotype + supplemented food
> C) mutant genotype + supplemented food
> Treatment/Genotype 	wt (W) 	mutant (M)
> Normal food (N) 	x 	NA
> Supplemented food (S) 	x 	x
> (x = data available, NA = not available)
> Each treatment was performed in four biological replicates, giving rise
> to 3 x 4 = 12 RNA samples.
> These 12 samples were analyzed on four 3-color microarrays,
> competitively hybridizing one sample from each treatment (A,B,C) to one
> array.
> Two contrasts are of interest to the researchers:
> 1.) For the wt genotype: genes with differential expression between the
> two food supplements (N, S)
> 2.) For "Supplemented food" (S) : genes with differential expression
> between wt and mutant genotypes (W, M)
> As these two question each refer to a single factor (either genotype OR
> food), I could perform two separate analyses on the data e.g. by
> treating the arrays like standard two-color hybridizations and
> extracting only the two channels of interest each time.
> Of course, I would be grateful for any advice,
> thanks,
> Thomas
> Wolfgang Huber wrote:
>> Hi Thomas
>> the NchannelSet class in the Biobase package can store such data [1],
>> some of the normalisation [2] and QC-assessment [3] methods that are
>> available for one- and two-color arrays can be either used directly or
>> with a little adaptation to such data, as can the linear model based
>> analysis of limma (e.g. by treating n 3-color arrays like 3n 1-color
>> arrays).
>> To be more specific, I think you will need to reveal the biological
>> question and the experimental design behind these data.
>> Best wishes
>> Wolfgang
>> [1] Have a look at the vignette of the CCl4 package "From the Genepix
>> data files to RGList to NChannelSet" for an example where such an
>> object is constructed, which you will need to adapt to the particular
>> file format your friend uses (you'll have to modify the read.images
>> function or emulate it with calls to read.table).
>> [2] vsn, quantiles, ...
>> [3] boxplots, MA-plots, between-array distance heatmap, such as in the
>> arrayQualityMetrics package
>> Il Oct/15/10 11:08 AM, Thomas Sandmann ha scritto:
>>> Dear all,
>>> I have received data obtained using a three-color microarray platform,
>>> e.g. three samples were labeled with three different fluorophores and
>>> hybridized competitively to a single array. Would anyone be able to
>>> point out a useful package for the analysis of three-color
>>> hybridizations ?
>>> Thanks a lot,
>>> Thomas
>> _______________________________________________
>> Bioconductor mailing list
>> Bioconductor at stat.math.ethz.ch
>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the archives:
>> http://news.gmane.org/gmane.science.biology.informatics.conductor

More information about the Bioconductor mailing list