Question: Meta analysis of healthy and diseased data using illumina microarrays
gravatar for chris86
2.5 years ago by
UCL, United Kingdom
chris86380 wrote:


I am in progress of trying to understand how to combine two separate microarray experiments run on the same Illumina platform, but several months apart, run in the same lab. The problem is, that healthy individuals and diseased individuals are in the two respective batches or experiments. From what I understand this makes any accurate batch correction impossible (?).

Further, if I want to combine them, I was thinking of quantile normalising separately, then I have read on another thread to try and POE normalise separately and then combine? However, if they are completely different experiments, I think POE normalisation - which converts gene by gene to 0 to 1 probability of expression is not the right method to use.

I was also thinking of using a rank method (rankprod) to do differential expression, I guess this makes more sense, but some genes could be DE due to a batch effect still.

Is there a best or better method here? Or is this idea just a bit crazy?

Please don't ask me why it was done like this, it depresses me to think about it.



ADD COMMENTlink modified 2.5 years ago by phil.chapman140 • written 2.5 years ago by chris86380
gravatar for phil.chapman
2.5 years ago by
United Kingdom
phil.chapman140 wrote:

I think in this situation the best approach is to normalise together and do the analysis, but when reporting the results back give a very clear caveat that there's a batch effect that it's impossible to correct for that may confound the results.  How much of an issue this is will depend on the size of the biological effect vs the amount of variability in the system: a large effect may well drown out the bacth effect anyway.  But if you end up with very modest effect sizes then the likelihood of the changes you see being down to the batch effect are increased.

You probably know all this anyway but that's what I would do, and also tell your colleagues to involve you in experimental design next time... :)

ADD COMMENTlink written 2.5 years ago by phil.chapman140
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 208 users visited in the last hour