Quantile normalization of microarray

Question

[tonja.r]

I am reading two papers of Bolstad about quantile normalization.

In "Probe Level Quantile Normalization of High Density Oligonucleotide Array Data" he writes about quantile normalization:

One problem with this method is that in the tails in particular,where we might expect greater differentiation between chips,the normalized values are going to be identical

However, in "A comparison of normalization nethods for high density oligonucleotide array data based on variance and bias" he writes:

This would be most problematic in the tails where it is possible that a probe could have the same value across all the arrays.

If I understood it correctly, in the first one he states that in tails (say where we have high intensities) the probe will have greater differentiation between the chips, so the values of the probe will not be the same between the chips.

In the second he states that a probe still could have the same value across the chips.

Isn`t it controversially? Or I understood the first paper wrongly and he is talking about the probeset in the first paper?

edit: or in the first paper he talks about unnormalized values and in the second about already normalized?

Thanks in advance.
Tonja

Answer 1

[Ahill]

He is referring to the same effect of quantile normalization in both sentences. When he says "greater differentiation" he means in reality, before quantile normalization. When he says "same values across all arrays" he is referring to post-normalization values. Consider an extreme case of a probe intensity for an interesting probe that moves to the 100th percentile in a small subset of samples representing a treatment group. After normalization, that probe's intensity in that subset of samples will be identical to the 100th percentile intensities of all the other arrays, potentially dampening an interesting change. Or, if a single probe (corresponding to a very abundant mRNA) was at the 100th percentile of all samples, after normalization it will have a single, identical, constant value across the entire dataset.