Normalization of oligonucleotide arrays /bolstad

Question

[tonja.r]

I am currently reading a paper of Bolstad "A comparison of normalization nethods for high density oligonucleotide array data based on variance and bias: and I have some questions about the result-section.

He writes:

Plotting the log of the ratio of variances versus the average of the log of the mean (expression measure across arrays)..

for example:

"the log of the ratio of variances" is then log[ Variance(Quantil) / Variance(unnormalized) ]
"the average of the log of the mean" is then 0.5 [log (Mean(Quantil) Mean(unnormalized) ]

Is it right? Because a few lines later he mentions: ...the first method ib the ratio has the smaller ratio...

has the smaller ration of what? reatio of variance(Qunatil) and mean(Quantil)? but we did not compare that.

Answer 1

[Ahill]

Yes, the "log of the ratio of the variances" is log[ Var(Quantile normalized RMA) / Var(unnormalized RMA) ].

The "average of the log mean" is the average RMA (not an average of quantile and unnormalized). This estimates the overall expression level of each probeset.

When he says "the first method in the ratio has the smaller ratio" he means log of the ratio of the variances (defined above). For example, in figure 4 (top left), the ratio quantile:unnormalized is less than zero at almost all intensities, indicating that quantile normalized RMA had lower variance compared to unnormalized RMA.