Bug running harmony batch correction with multiple covariates

[joseverdezoto]

Thanks for making this awesome package! I'm having some problems when running sc.pp.harmony. I previously had v2.7.0 installed and ran into an issue when defining multiple covariates in the batch argument. I updated the package installation to v2.7.1 and the function works as expected with a single covariate. However, if I provide a list of covariates, I get the following error:

My coworker and I tracked down the problem to trying to create a pandas df with pd.DataFrame(adata.obs[batch]). If you simply run pd.DataFrame(adata.obs) without using the slicing operator to retrieve specific columns, the df is created properly. We also tried pd.DataFrame(adata.obs[:][batch]) and the df with the batch covariates is properly created. Not sure if we're right, but we think this is related to snapatac2 trying to use a reference to the actual data for making the covariate df.

Would love to hear your thoughts. Thanks so much for your help!

[jpmcga]

I've encountered the same error when trying to access multiple .obs columns on a backed SnapATAC2.AnnData/AnnDataSet object. Sorry if I'm stating what you already know, but I think this is due to the SnapATAC2's 'out-of-core' implementation of AnnData objects, wherein .obs is a 'DataFrame element' not a pandas DataFrame.

See https://kzhang.org/epigenomics-analysis/anndata.html.