What Are The Downsides Of Replicated Spots On A Microarray?

Question

[Eric Fournier]

Greetings,

we are currently in the process of designing a microarray to be used in experiments measuring the level of methylation of the genome. One possible design we are considering involves printing all probes in pair with their reverse complement to be used as either a control or a way of assaying the levels of hemimethylation, depending on the exact nature of the sample processing.

However, I am starting to get concerned that such "almost replicates" spot will become a terrible sore in downstream analysis. I'm currently reviewing the relevant literature (such as http://www.hindawi.com/journals/cfg/2009/950171/) but would appreciate input from the Biostar community regarding downfalls I might be ignoring.

[seidel]

Will the labeled material be single or double stranded? (i.e. just making sure your labeling strategy really is strand specific). Why would these spots be a terrible sore? You could always just ignore them in the data analysis, no? Also, they're not really replicated spots (as your title suggests), they are different, reverse compliment spots. What are they actually controlling for? Is it some very predictable property? If so, then I think they would be worth having. If you're unsure what they're controlling for, they would likely cause confusion.

[Eric Fournier]

Both single and double stranded material will be used in separate experiments (the microarray will be part of a platform used by different researchers and students).

I'm mostly worried that the "almost-replicate" spots might be problematic by breaking some underlying statistical assumption used in microarray normalization and/or differential analysis.

In experiments involving amplification (double-stranded material), the reverse-complement would be used to control for false positives. In experiments without amplification, they would be sued to measure hemimethylation.

Answer 1

[Aaron Statham]

I've used sense/antisense probe pairs on a Nimblegen custom array and when double stranded labelled material was hybed to them, the signals from the probe pairs indeed are highly correlated (pearson r>0.8). Unless you actually wish to interrogate each strand separately (therefore need single stranded labelled material) I wouldn't bother with the sense/antisense pairs.

Answer 2

[Stefano Berri]

I can't see any. You will have twice as many dataponts. You need to take the average between the technical replicate (sense and anti-sense), I think LIMMA does it for you already. This will reduce the noise across samples increasing the power of your tests. If the spotting has little extra cost, you could spot 4 or more replicates (spatially distant) to further reduce noise.