[BioC] Correcting for batch effects in limma

[James W. MacDonald]

Hi Khadeeja,

On 5/15/2012 6:06 AM, khadeeja ismail wrote:
> Hi,
>
> Is it possible to correct for batch effects in limma when doing a paired analysis? I have pairs from two
> runs Batch 1 and Batch 2. There  are no pairs where one is in Batch1 and the
> other in Batch 2. If I enter the Batch no. into the design matrix,
> no coefficients are generated as there is no difference between run no.
> for any pair.

Without seeing your code it is hard to say much. In addition, it isn't 
really clear what you mean by 'no coefficients are generated'.  However, 
you should note that accounting for the pairing structure and the 
batches will to a certain extent be doing the same thing. As an example, 
let's consider a single pair from one batch.

If we were to consider the conventional approach for paired data, you 
would first compute pair1_treated -  pair1_control, and then using these 
differences to compute statistics. By computing the paired differences, 
we have subtracted out any sample-specific variability, which includes a 
batch effect (e.g., if batch 1 has higher overall expression due to some 
technical reasons, you would expect both of the pairs to reflect this 
higher expression, and subtracting the two would thus eliminate the 
batch effect, modulo variability).

When you fit a batch effect, you are in essence computing a mean 
expression value for all samples in a particular batch and then 
subtracting that from each sample. This is very similar to what you have 
done by pairing. Doing both is not likely IMO to add benefit, and is 
just wasting degrees of freedom.

Best,

Jim


>
> Any advice on this would be most appreciated.
>
> Thanks,
> Khadeeja
>
> 	[[alternative HTML version deleted]]
>
>
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor

-- 
James W. MacDonald, M.S.
Biostatistician
University of Washington
Environmental and Occupational Health Sciences
4225 Roosevelt Way NE, # 100
Seattle WA 98105-6099