Issue with CCC edges only within spots when using deconvoluted data (Cytospace output: same coordinates for multiple cells) #26
Description
Hello CellNEST developers,
Thank you for developing this powerful tool. I encountered an issue when using CellNEST with input data derived from Cytospace deconvolution of Visium data.
Problem:
All predicted CCC edges are confined within individual spots, and no inter-spot connections are found, unlike the cross-spot arrows shown in your examples.
My Input Data:
After deconvolution with Cytospace, each Visium spot contains multiple cells (~ 5 cells). These cells from the same spot share the exact same original x, y coordinates. I suspect this is causing issues in the distance calculation and graph construction.
My plot:
What I've Tried:
1.I manually calculated the distances between cells and found that the smallest non-zero distances are clustered around 355-357 μm, with the next level around that value multiplied by ~1.732.
2.Based on this, I adjusted juxtacrine_distanceand neighborhood_thresholdparameters (e.g., 400, 700) to try to capture inter-spot connections.
3.I also tried using the KNN method (k=50) for neighborhood definition.
4.Unfortunately, none of these attempts successfully produced CCC edges between different spots.
My Hypothesis:
The fact that numerous cells share identical coordinates might be causing the distance matrix to be dominated by zero distances and these very small non-zero values. This could potentially lead the GAT to assign disproportionately high attention/weights to intra-spot edges, making it difficult to detect the weaker but biologically relevant inter-spot CCC signals.
Question:
Could you provide any guidance on how to properly preprocess or handle such deconvoluted data where multiple cells share the same initial coordinates? Are there specific strategies or parameter tuning tips to mitigate this issue and allow CellNEST to detect inter-spot communications?
Thank you for your time and support.
Best regards!