Inquiry on Neighbor Graph Methods for 8 µm Bin Data in Visium HD

[zhenhuapku]

Description of feature

I hope this message finds you well. I am currently analyzing spatial transcriptomics data from Visium HD, specifically focusing on the 8 µm bin data, and am considering different approaches for constructing spatial neighbor graphs to explore cell interactions within the tissue microenvironment.

For my analysis, I am evaluating three methods for creating these neighbor graphs:

Grid Mode: Utilizing the fixed grid coordinates associated with the Visium platform.
20-Neighbor Mode: Calculating connections based on the 20 nearest cells.
Delaunay Triangulation: Constructing the graph based on Delaunay triangulation.
I would greatly appreciate your insights on the following:

Which of these methods do you recommend for analyzing 8 µm bin data from Visium HD, considering aspects like resolution and spatial accuracy?
What are the advantages and limitations of each approach in this specific context?
Are there best practices or specific considerations I should keep in mind when applying these methods?
Your expertise would be extremely helpful in guiding my analysis strategy.

Thank you very much for your assistance!

[marcovarrone]

Since you are using the grid spots as "cells" I suggest to use the grid mode.
If with grid you mean taking the 8 neighbors surrounding a spot, I predict that grid mode and Delaunay triangulation will give very similar results (except maybe at the borders).

I would suggest the 20 neighbors mode only in case you want to consider niches that consider groups of cells that could be very far away from each other. But in that case, I would still suggest the grid mode while increasing the number of neighborhood layers in CellCharter (but this comes at a cost of higher complexity).