Strategy needed: RCTD deconvolution for large-scale mouse liver spatial transcriptomics data (99,894 spots)

[Lao-Tz]

I'm analyzing spatial transcriptomics data from mouse liver in Type 2 Diabetes, with matching scRNA-seq reference data. The dataset contains 99,894 spots (52,006 features), stored in a Seurat object:

st_sp1
An object of class Seurat
52006 features across 99894 samples within 3 assays
Active assay: integrated (3000 features, 3000 variable features)
2 layers present: data, scale.data
2 other assays present: Spatial, SCT
2 dimensional reductions calculated: pca, umap
1 image present: sample1

I'm facing two main issues:

1. Visualization problems:

• The number of spots is too large to generate visualization plots
• Even when plots are generated, they're too dense to interpret meaningfully

2. RCTD deconvolution results appear problematic:
   Using 'doublet' mode, the results show:
   First type assignment:

• Hepatocytes: 99,892 spots
• Cholangiocytes: 2 spots
• All other cell types: 0 spots

Second type assignment:

• Cholangiocytes: 85,928 spots
• Kupffer cells: 6,899 spots
• Macrophages: 2,961 spots
• Other cell types: (various smaller numbers)

Questions:

1. Should I:
   • Re-run RCTD with 'full' mode instead of 'doublet'?
   • Merge adjacent spots before deconvolution?
   • Try a different approach entirely?
2. Are these results typical for liver spatial data, given the high proportion of hepatocytes?
3. What's the recommended approach for handling such a large number of spots while maintaining biological relevance?

Technical details:

• Platform: 10x Visium
• Sample: Mouse liver (T2D model)
• Analysis: RCTD deconvolution with matching scRNA-seq reference
• Current parameters: doublet mode

Any suggestions would be greatly appreciated!