Running NiCo on an integrated object

[reinertanalytics]

Hello, thanks so much for an excellent package!! We've really enjoyed using it on our xenium data.

Would it be possible to run NiCo on an integrated object? My concern is that the preprocessing steps apply SCTransform to the shared gene space (requiring raw counts), but our integrated object has already been log transformed. Would we just need to go in and add the raw counts to the integrated object, or is there another way to do this?

[ankitbioinfo]

Hello, thank you for trying NiCo, and I am glad to hear that you enjoyed running it on your xenium data!

Regarding your question, could you clarify what you mean by "integrated object"? Are you referring to multiple tissue slides stored in a spatial Anndata object?

In general, we felt applying scTransform to a shared gene space is not stable for preprocessing. So, in the final version of NiCo tutorial, SCTransform is applied to the full gene space for both alternative preprocessing methods. Here's how it was done:

#Copying the full gene space sequencing AnnData object 
ad_seq_common=ad_seq_ori.copy()

#Alternative 1: Using scanpy version of normalization (Recommeded)  
sc.experimental.pp.normalize_pearson_residuals(ad_seq_common,inplace=True)

#Alternative 2: External script with scTransform normalization  
sct_ad_sc = sann.SCTransform(ad_seq_common,min_cells=1,gmean_eps=1,n_genes=500,n_cells=None, #use all cells
		                bin_size=500,bw_adjust=3,inplace=False)

Later, when you run the find_anchor_cells_between_ref_and_query function, the Annotation module will operate on the shared gene space by default, and the next integrated NMF in the covariation step will also proceed similarly in the shared gene space.

If you already have other normalized version of counts then just skip the scTransform normalization and proceed with remaining commands. I hope that should work.

I hope this helps! Feel free to reach out with further questions or clarifications.

[reinertanalytics]

It did work on my integrated object, thanks for your response!