scanorama from this repo vs scanpy external

[mortunco]

Hi,

Thank you very much for the tool. I am very interested using this tool as the only label we have in the data is the dataset names (~samples). The remaining labels we have is somewhat arbitrary. I would like to try batch correction to eliminate the dataset differences. I am aware that our biases are complex such as different cancer + different lab, but I just wanna give a try. I apologise if its too much stuff to correct.

I have a dataset of normalized values. I previously merged this dataset into a single anndata object. We use "dataset" column for the batch labeling. In this data I have 27k cells.

I am little bit confused about doing batch corrections with scanorama.

method 1

Using https://scanpy.readthedocs.io/en/stable/generated/scanpy.external.pp.scanorama_integrate.html#

Load the object.
Normalize and PCA.
Add batch measure.
Integrate.

import scanpy as sc
import scanpy.external as sce
adata = sc.datasets.pbmc3k()
sc.pp.recipe_zheng17(adata)
sc.tl.pca(adata)
adata.obs['batch'] = 1350*['a'] + 1350*['b']
sce.pp.scanorama_integrate(adata, 'batch')
'X_scanorama' in adata.obsm

method 2

https://github.com/brianhie/scanorama#troubleshooting
Join all adatas into a list.
Integrate. Batch correct

# List of datasets:
adatas = [ list of scanpy.AnnData ]
import scanorama
# Integration.
scanorama.integrate_scanpy(adatas)
# Batch correction.
corrected = scanorama.correct_scanpy(adatas)
# Integration and batch correction.
corrected = scanorama.correct_scanpy(adatas, return_dimred=True)

Question 1.

• In the tutorial for method2, I always see creating the list of AnnDatas. Given I have already have my cells in a single anndata object do I have to split it again?
• Method2 does not have a batch key different than sce.pp.scanorama_integrate(adata, 'batch') . Should I be worried about this?
• Method1 does not have correction? Should I add correct_scanpy to Method1?

Thank you very much for your help,

Best regards,

Tunc.