QuantileNorm :: Probe-level quantile normalization module

NimbleGen microarray system applies in situ oligonucleotide synthesis technology and usually produces hybridization data of single fluorescence channel, which makes the normalization procedure much easier than other platform such as DNA array and Oligo array that needs to remove the print-tip variation and dye-bias during the processing. The only thing we need to take care is to remove the variations resulted from multiple microarray experiments by which quantile normalization algorithm is strongly recommended.This module in Nimble provides two ways to carry out the procedure, namely the one-step quantile normalization and the two-step quantile normalization. The one-step way is to normalize all the chips at one time, enforcing the distributions of all the slides to be the absolutely same, which is a more robust method that removes all the variations during replicate experiments and the differences from the RNA sample preparation, but at the expense that some of real biological differences might be smoothed to a certain of content. The latter method divides the procedure into two steps: first, perform quantile normalization on replicate slides within each tissue, and then perform global normalization to adjust all tissues to an average baseline.

How does the user to choose a two-step normalization or a one-step normalization

NMPP provide the option for users to choose a two-step normalization procedure and a one-step procedure, the users need to make their own decision according to their own experiment design. Here is our suggestion:

Quantile normalization forces the arrays in a set of experiments to have absolutely identical distribution, based on the assumption that the RNA populations hybridized to the arrays should be the same. However, the content and quantity of RNA transcripts in different tissues usually vary a lot, and rash quantile normalization across all the chips might be likely to conceal or weaken the real biological differential expressions. So, when the user is working on a sereis of tissues-specific arrays, for drawing a developmental map, two-step normalization is suggested. and if the user using microarray to screen the mutant genes, such as a wt/mutant comparison experiment during one tissue, one-step quantile normalization is suitable for a strict screening of the target genes.

from the above clustering tree based on the normalized data by two-step procedure, we can see that the tissues in the same organ have a close relationship distance. But by one-step procdedure, we failed to get a proper relationship of tissues, since the real biological difference has been improperly smoothed.