[BioC] ComBat - question about normalization method

[Johnson, William Evan]

Maren, 

I hope you don't mind but I cc'd the bioconductor mailing list as well. 

In most cases, I think ComBat should work okay on the pre-normalized data. However, I can imagine cases where it may not work as well, so you may have to be careful. For example, if you have one batch that has more low to moderate quality samples than another, then the batch effects could reappear once you do the normalization or downstream analysis--depending on what you do (I am not familiar with your S-Score method, so maybe this isn't a problem). 

However, what I would strongly recommend to you is to use our SCAN normalization approach (Piccolo et al. 2012, Genomics--Bioconductor: 'SCAN.UPC'), that does a probe-level, single-sample normalization, followed by ComBat (Bioconductor: 'sva'). This will background correct, normalize, and adjust for batch all on the probe level. Then you are all set for downstream analysis.

Hope this helps,

Evan


On Apr 22, 2013, at 2:13 PM, Maren Smith wrote:

> Dr. Johnson,
> 
> I am a graduate student at Dr. Michael Miles laboratory at Virginia Commonwealth University.
> 
> I have a question about the usage of the ComBat script. Normally ComBat is used after RMA correction
> However, for our microarray data we would like to use the ComBat script on the raw expression data to control for background effects and then use the S-Score method developed at this laboratory to identify changes in gene expression between control and experimental groups.
> In your opinion, is this an appropriate use of ComBat?
> 
> Thank you for any insight you can provide,
> Maren Smith.