Microarray batch effect correction

Question

[sarahmanderni]

Hi,

I intend to integrate two microarray datasets with imbalanced outcome. First of all which method has shown the most robust results in this context. My current options are ComBat and RUV-2.

Second, I searched on how to use ComBat. I found this article highlighting the fact that taking into account the study outcome as a covariate will exaggerate the confidence of downstream analysis. On the other hand there is this post by the package author which suggests using the outcome as a covariate in addition to the batches. Now I am not sure which one to choose.

Thanks in advance!

Answer 1

[Kevin Blighe]

Dear Sarah,

There have been a few recent posts regarding batch correction (generally) and ComBat. Please take a look at the suggestions in this thread: RNA-Seq Batch correction negative values

I'll also direct you to a great read on this topic: https://www.ncbi.nlm.nih.gov/pubmed/20838408

As your data is microarray, I would favour dividing your dataset into training and validation instead of trying to 'neutralise' the batch effect, assuming that your sample size is good of course.

Kevin