liftCellChat() corrupts centrality mapping (netP$centr): sender identities swap in signalingRole_scatter

[mhvbyrne]

I am observing a reproducible issue where liftCellChat() causes misalignment of centrality values (netP$centr) with cell type labels, leading to incorrect sender/receiver identification in downstream plots such as:

netAnalysis_signalingRole_scatter()

Summary of the problem
CellChat objects behave correctly before lift
After liftCellChat(), cell labels (@idents) remain correct
However, centrality values (outdeg, indeg) are reassigned to different cell types
This results in incorrect biological interpretation

obj_a@idents and obj_b@idents both have additional cell types which the other doesn't have

Example of the issue
group.new <- union(levels(obj_a@idents), levels(obj_b@idents))

obj_b_lift <- liftCellChat(obj_b, group.new)

targets <- c("pericyte", "pericyte_STEAP4", "fibro_CXCL14")

p <- names(obj_b@netP$centr)[1]

data.frame(
cell_type = targets,
outdeg_before = obj_b@netP$centr[[p]]$outdeg[targets],
outdeg_after = obj_b_lift@netP$centr[[p]]$outdeg[targets]
)

Output:

    cell_type outdeg_before outdeg_after

pericyte 0.1270253 0.0010619
pericyte_STEAP4 0.0005287 0.1270253

The values are effectively swapped between some cell types.

Additional diagnostics
identical(
names(obj_b@netP$centr[[p]]$outdeg),
names(obj_b_lift@netP$centr[[p]]$outdeg)
)

Result:
FALSE

So the issue appears to be a mismatch between centrality vector ordering and cell identity levels after lift.

Impact
This directly affects:
netAnalysis_signalingRole_scatter()
Any analysis using netP$centr
Interpretation of major signaling senders/receivers

CellChat v2.2.0

I ran netAnalysis_computeCentrality on my original cellchat objects, and the cell labels were stored as factors originally
I have tried recomputing NetP before and after lift and it still doesn't fix the issue:
cellchat_obj_a <- netAnalysis_computeCentrality(cellchat_obj_a, slot.name = "netP")
cellchat_obj_b <- netAnalysis_computeCentrality(cellchat_obj_b, slot.name = "netP")

Please can you advise?

[mhvbyrne]

I have resolved this issue by not using the lift function, and instead taking an intersection of the labels of the two seurat objects, then rerunning the make cellchat pipeline for each of them, and then performing merge without the lift function.