NiCo on multilple sections

[kewinds]

Hey Ankit,
First, thank you for this wonderful tool! We've gotten great results. I have multiple sections from two different disease conditions that I'd like to perform this analysis on. Do you have any recommendations for how I could implement your pipeline? I'm especially interested in overall trends in disease vs control tissue, which would require somehow integrating all of the analyses together.

[ankitbioinfo]

Hello @kewinds,

I am glad to hear that you have found some nice results with NiCo. Your question about how to include multiple sections in one unified analysis is great. To do this, you have to do some tricks. Since multiple sections can have local coordinate frames, which can overlap if you combine them directly. You can shift the XY coordinate (X+shift, Y+shift) values of your sample 2 in comparison to sample 1. This way, it will have one unified global coordinate and multiple sections. The value of shift has to be figured out according to your local tissue sections' coordinate values. You can see in the example image below two tissue section samples can be analyzed combined with NiCo.
I hope this will help.

[kewinds]

Thanks so much for the quick response! A few more follow up questions for you:

1. I have a total of 8 samples on which I've individually run NiCo. I could foresee trying to combine all that data would be computationally intense (we have about 1 million cells). I am pretty happy with the individual cell annotations that I already have, so I'm wondering if there's anyway to somehow "average" all of those analyses together. Unless you think NiCo can handle it.
2. Do I need to integrate the samples at all to correct for batch effects? If so, any suggestions for packages that would be compatible with this analysis?

[ankitbioinfo]

Hi @kewinds,

Earlier, I ran the full pipeline on ~0.4 million cells using a standard laptop, and it completed successfully. Since you mentioned you're satisfied with the individual sample cell annotations (which is the most time-consuming step), you can still proceed with the niche interaction and covariation analyses on the fully integrated dataset.
To do this, you can merge all eight samples while ensuring that the spatial XY coordinates from different samples do not overlap. Then, append your existing annotations to the combined dataset and replace the files "nico_celltype_annotation.h5ad" and "sct_spatial.h5ad"
After that, you can re-run NiCo starting from the spatial_neighborhood_analysis step, skipping all the prior annotation steps. This approach should allow you to obtain an integrated analysis across all samples.
If all the samples come from the same experimental protocol and spatial technology, batch effects are usually minimal. However, I’d still recommend performing basic exploratory analyses—like comparing average gene counts, total UMI per cell, or PCA plots across samples—to confirm consistency.

If you suspect batch effects or want to be cautious, you could consider using Python-based tools like:
scVI or scANVI for integrated modeling of spatial and single-cell transcriptomic data.
Harmony or Scanorama (available through scanpy ecosystem) for batch effect correction, though they are primarily designed for scRNA-seq and may require adaptation for spatial data.

Let me know if you'd any more question.

[ankitbioinfo]

I discover one more solution for using multiple sections. You can add artificially 3rd dimension Z for every section with specified interval. Since the NiCo default distance cutoff do not account any neighbors which is really far (100). So the adding Z dimension for interval of 200 could be an optimal.

PID=np.unique(ad.obs['Patient_ID'])
n=len(ad.obs_names)
Zcoord= np.zeros((n,3),dtype=float)

count=1000
for i in range(len(PID)):
    ind=np.where(PID[i]==ad.obs['Patient_ID'])[0]
    Zcoord[ind,0]= ad.obsm['spatial'][ind,0]
    Zcoord[ind,1]= ad.obsm['spatial'][ind,1]
    Zcoord[ind,2]= count
    count=count+200

ad.obsm['spatial2']=Zcoord

Run the neighborhood analysis using 'spatial2' in following way.

niche_pred_output=sint.spatial_neighborhood_analysis(
Radius=inputRadius,
output_nico_dir=output_nico_dir,
anndata_object_name=annotation_save_fname,
spatial_cluster_tag=annotation_slot,
spatial_coordinate_tag='spatial2',
removed_CTs_before_finding_CT_CT_interactions=do_not_use_following_CT_in_niche)

[kewinds]

Thanks so much for the suggestion! Will give it a try. Your initial solution worked very well also.