Correct more than two batches

[parkjooyoung99]

Hello,
I am running fastMNN with data set that having 2 batches , 'dataset' and 'condition' .

I tried to correct both in once by specifying batch
fastMNN(combined,batch=c(combined$dataset, combined$condition), subset.row=chosen.hvgs)
but this does not work.

Is there a way to correct multiple batches?

Thank you :)

[LTLA]

I assume you mean you have two factors, of which each combination of levels is to be treated as a batch. If so, just paste the factors together. c() doesn't make sense here.

[parkjooyoung99]

@LTLA

I tried that but that actually did not correct batches.
I guess the following is happening,

Let's assume we have 4 datasets, A.B.C.D, where A and B are consist of patients and the others normal.
What I expected is to correct differences A vs B vsC vs D and plus (A,B) vs (C,D).
However, when I merge them as A_patient, B_patient, C_normal, D_normal,
fastmnn only corrects differences A_patient vs B_patient vs C_normal vs D_normal, but not (A_patient, B_patient) vs (C_normal, D_normal).

Hope that there will be an update about this!

Thank you!

[LTLA]

I don't understand your issue. If fastMNN has removed batch effects between the individual datasets, then why is the (A, B) vs (C, D) still present? If A == B == C == D, then this implies that (A, B) == (C, D); what makes you think otherwise?

[parkjooyoung99]

Batch between individual datasets and batch between patient vs normal is a different story.

When correcting batches between datasets, differences among 4 datasets should be considered.
However, to correct the batch between patient/normal, it has to only consider the differences between (patients vs normal), not the differences between individuals.

When merging two batches as ["patient, "__normal],
fastmnn does not think about differences between patient vs normal, since it recognizes each "" as one batch level.
To say in other words, A_patient and B_patient should be recognized as the same level (patient) in the context of patient/normal but fastmnn does not consider them as the same level.

Indeed, it did not correct patient/normal batch.

I hope it clarifies my idea.!

[LTLA]

If you don't want to correct the differences between A and B, then just set the "patient/normal" factor as your batch=, in which case the differences within the patient level (i.e., between individuals) will be ignored. From experience with human data, I don't think this is a good idea, but whatever floats your boat.

If this is not what you are referring to, then I would suggest preparing a minimum reproducible example before proceeding with any further discussion.