snRNAseq data integration for annotation reference

[gomezdj]

Can we create an integration solution for Annotations with sn-RNAseq (reference) that can search for snRNAseq neighbors and spatial neighbors? Does the same set of files need to be used for the inputRef and inputQuery? Does the SCTransform improve the results for final output sc-spatial data. I know the gene names have to be the same in both datasets (snRNAseq and spatial) but how do we ensure the shape is preserved?

Thanks,

Daniel Joseph Gomez

[ankitbioinfo]

Hi Daniel,

Thank you very much for trying our tool NiCo and sharing your notebooks. I have run NiCo with snRNAseq, and it worked flawlessly. To get good annotations on any spatial data is tricky, so a lot of parameters need to be explored. So here is my answer to your queries.

(1) Yes, NiCo works for snRNA-seq as well. The gene and cell filter criteria might need to be adjusted for better mapping between reference and query data. In your spatial data, if some cell type morphology is funny and quite large, then such cells' transcripts might not be able to map with nuclei transcripts from snRNAseq. In the majority of cases, cell type annotations should be fine.

(2) The different files and anndata layer as follows for NiCo analysis.

Original_counts.h5ad (.raw.X contains raw counts, .X contains any normalized counts)
sct_singleCell.h5ad (.raw.X contains raw counts, .X contains sct normalized counts)
sct_spatial.h5ad (.raw.X contains raw counts, .X contains sct normalized counts)

.raw.X counts are required for the Covariation step, and .X sct counts are needed for the Annotation step, .X any normalized count is needed for just visualization purposes.

If I understand your notebook correctly, then you have not performed the SCT normalization. And I do not know how many genes overlap between the reference and query. I need some more details, such as the output of this print in your datasets, to understand this error "ValueError: Found array with 0 feature(s) (shape=(2239, 0)) while a minimum of 2 is required by TruncatedSVD".

print(len(sp_genename[index_sp]))
print(len(sc_genename[index_sc]))

or  len(set(ref_adata.var_names).intersection(query_adata.var_names))

(3) scTransform is only needed for the Annotation step, neither for the Interaction nor for the Covariation step. I see in your spatial data that 'CellType' is already present in the obs layer. If you are not interested in spatial data to annotate it again, then you can skip the annotation task and directly start with Interaction with the following module sint.spatial_neighborhood_analysis.
You can also look at another tutorial nico_analysis_lowres_seq_tech.ipynb that basically starts with the Interaction step directly.

(4) If you want to apply the Covariation module for the full gene dataset of snRNA-seq, then your cell type names between spatial annotation and snRNA-seq annotation must be identical.

(5) Please let me know how many genes overlap between the query and reference, then I can further understand the error in your shape mismatch.

Thanks, and please do not hesitate to ask any further questions.