Normalising Arrays Across Multiple Datasets For Unsupervised Clustering.

Question

[Ankur]

Hi there!

I'm interested in using multiple datasets consisting of arrays from the same type of cancer to check for the presence of molecular subtypes. What is the best way by which I can normalize all the arrays so they become comparable before I run consensus clustering? Combat + rma?

Cheers, Ankur.

Answer 1

[brentp]

This is a difficult topic and you'll find some previous answers searching biostar (e.g. Microarray meta-analysis)

There is an R-package that implements some methods of combining datasets: http://cran.r-project.org/web/packages/MAMA/index.html It is less focused on normalization, and more on combining via p-values, effect-sizes, or order.

There are also quite a few papers using some of these methods, e.g.: http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0045506

[Ankur]

I've used microarray meta-analysis before in a different setting but the fundamental caveat is that those tend to be supervised analyses, with the P.values et cetera being generated in relation to normal tissue, for instance, whereas I'm looking to use unsupervised clustering here in this case to discover previously unknown tumour subtypes across a single type of tumour, which necessitates making arrays directly comparable across studies...

Cheers.

Answer 2

[Ankur]

I did find a possible solution though - given that my arrays are all of a single Affymetrix platform I will be able to apply frozen RMA normalisation to facilitate direct comparison.