Can we use previous estimated ComBat model to correct batch effects for new data?

Question

[Min Dai]

Hi,

I am puzzled about this situation: if we have estimated a ComBat model for batch effects in dataset A, can we use this model to adjust batch effect for dataset B?

In another case, there are two batches, i.e., batch 1 and batch 2, in your dataset, but you just want to use 70% data of high quality to estimate and remove batch effects for batch 2 (if we set batch 1 as the reference batch). After estimating the model, we also need to adjust batch effects for the rest 30% data of batch 2, how can we do this?

Thanks in advance,
BP

Answer 1

[Kevin Blighe]

I would question this approach, as it assumes that the batch effects are the same in the new data - this is a big assumption to make, though. If at all possible, just include batch as a covariate in your design formula(e) while making statistical inferences.

If you must directly remove batch somewhere, then use removeBatchEffect() from limma.