Batch effect correction

Question

[gstavropoulos1991]

Dear all,

I want to correct for batch effects. I have two data sets, two similar diseases, measured at completely different time points. When I do PCA, I clearly see two groups; however, this shouldn't be the case since those two diseases are rather similar and therefore, that's a clear indication of batch effects.

The thing is that I'm just getting started with batch effect corrections and therefore, I'm not really familiar with how I should properly do the analysis step by step. Consequently, I would appreciate if someone could shed some light on how I should proceed.

By the way, I have already read the SVA paper by Leek et al. but I didn't really get into the topic or how I should proceed.

Thank you in advance.

Regards, George.

Answer 1

[Kevin Blighe]

Presumably you have gene expression data? How did you process it? Why do you have a matlab tag for your question?

Generally there are 2 options that allow us to deal with batch:

1. include batch as a covariate in your design model (this will adjust the statistical values like any other covariate)
2. directly model batch and modify your raw counts prior to normalisation