search

ChaozhongLiu / MinNet

1 stars 0 forks source link

GLUE Batch Effect Removal #2

Closed valentinwust closed 1 year ago

valentinwust commented 1 year ago

In Fig.3 A and B you show pictures where GLUE doesn't properly remove batch effects. However, when I try to reproduce this with the same data the batch effect removal seems to works fine. Are you sure you have told GLUE the batch labels? I can't find the code for the model that was trained for Fig.3 A, but for example in this code for Fig.2 GLUE doesn't actually get the batch labels. As explained in their tutorial, this requires adding use_batch, like:

`scglue.models.configure_dataset(
    rna, "NB", use_highly_variable=True,
    use_layer="raw", use_rep="X_pca",
    use_batch="batch" # This does the batch effect removal
)
ChaozhongLiu commented 1 year ago

Hi,

Thanks for raising this up! First of all, I believe this is due to the update in GLUE. I followed the same website tutorial at the time when doing the benchmark, and I'm quite sure that the batch function doesn't exit at that time (early 2022).

I'm a big fan of GLUE and I'm excited to see it can do batch correction now (as I trust your results)

Let me know if you have any concerns!

Best, Chaozhong

valentinwust commented 1 year ago

Interesting, looks like you are right; I hadn't looked at the development history of GLUE. Batch correction made it into the final paper so I assumed this was always implemented, but it is indeed missing from the preprint.