Zero value in the computeRegionDistance

[JinAshley]

computeCommunProb fails for Xenium spatial data because computeRegionDistance returns zero group-level distances

Hi CellChat team,

I am running CellChat on Xenium single-cell spatial transcriptomics data. Xenium coordinates are already in microns, so I set ratio = 1. Following the guidance for single-cell spatial data, I tested small tol values and also tried tol = min center-to-center distance / 2.

However, computeCommunProb() fails when distance.use = TRUE because the internal group-level spatial distance matrix returned by computeRegionDistance() contains many zero values.

Data setup

I create the CellChat object from a normalized expression matrix:

cellchat <- createCellChat(
  object = data.input,
  meta = meta,
  group.by = "labels",
  datatype = "spatial",
  coordinates = spatial.locs,
  spatial.factors = spatial.factors
)

spatial.factors <- data.frame(
  ratio = 1,
  tol = 0.04
)

min(d.spatial[d.spatial != 0])
# 0.08259252

cellchat <- computeCommunProb(
  cellchat,
  type = "truncatedMean",
  trim = 0.1,
  distance.use = TRUE,
  interaction.range = 100,
  scale.distance = 1,
  contact.dependent = TRUE,
  contact.range = 10
)
### 
I get a error when run the computeCommuProb:
The suggested minimum value of scaled distances is in [1,2],
and the calculated value here is 0

Error:
Please increase the value of `scale.distance` and use a value that is slighly smaller than Inf



I wrapped computeRegionDistance() to inspect its output. It returns a list:
names(last_region_distance)
# "d.spatial" "adj.contact"
The returned d.spatial is group-level, not cell-level:
dim(last_region_distance$d.spatial)
# 9 9
The matrix contains many zeros:
Number of zero values: 71
So the failure appears to come from the zero values in the group-level d.spatial matrix being included in the minimum scaled-distance check. But I have checked that there would be no overlapping cell centroid coordinates,
Question
For Xenium/single-cell spatial data, what is the recommended way to handle zero values in the group-level d.spatial matrix returned by computeRegionDistance()?
Should off-diagonal zero values be interpreted as:
true spatial overlap / extremely close groups,
no valid/proximal group-level distance after filtering,
same-group/self-distance behavior,
or something else?
Would it be appropriate for CellChat to ignore diagonal zeros and/or ignore zero values when checking the minimum scaled distance?