Hi,

I've read capter 12 of the 2014 limma manual in order to identify DE genes in multiple arrays. There is one step where I'm not sure whether I did the right normalization.

In Cap. 12 the 'Aquantile' normalization between arrays is recommended:

> MA <- normalizeBetweenArrays(MA, method="Aquantile")

However, if I apply this to my 8-chamber Agilent data, I obtain a graph where the individual traces do not fit quite nicely.

I have found at Normalization for use of individual channels from 2-color arrays? that a combination of normalizeWithinArrays and normalizeBetweenArrays can be used instead. I used this:

MA.p <- normalizeWithinArrays(RG, method="loess")
MA.Aq <- normalizeBetweenArrays(MA.p, method="Aquantile")

plotDensitiesMA.Aq) now results in almost perfectly matching traces. Could anybody comment on this? Is this allowed?

Thanks for your replies!

Alfons

The relevant chapter of the limma User's Guide is Chapter 6 "Pre-Processing Two-Color Data". Chapter 12 covers the more advanced topic of single channel analysis.

For most analyses you need only the normalizeWithinArrays() step. The Aquantile normalization is only needed for single channel analysis. When normalizeBetweenArrays() is mentioned in Chapter 12, it is assumed that the normalizeWithinArrays() step has already been done. Aquantile normalization is never used by itself. If you are not doing a single channel analysis, then it will make no difference to the final results whether or not you do the normalizeBetweenArrays() step. 

In summary, the normalizeWithinArrays(RG, method="loess") is what you need and usually it is all you need.

Sean Davis gave the same answer to the earlier post 4 years ago.

Not only the densities will match. Do a boxplot to see that with this double normalization, the means and quantiles are normalized as well

This is the way to do it