What Parameters Should Be Considered When Introducing Noise In Microarray Expression Data?

Question

[Agatha]

If we consider a list of genes and their associated expression levels (fold changes), one would most probably tempted to introduce as noise variable the mean or the median (considering that the expression levels are part of a gaussian distribution).

What other parameters should be considered when trying to build a more accurate model ?

Answer 1

[Istvan Albert]

What I think you mean is to consider the distribution all genes as a background signal and look whether your selected genes could have been sampled by chance from this distribution.

You could also pick house-keeping genes as background, spiked in controls, or genes that you know don't/shouldn't change.

But the real/proper solution is replication, then you don't need to estimate noise levels from the other datapoints, instead you can estimate variability across replicates.