Cell-cell Communication with CellChat

Infer ligand-receptor networks from a single-cell dataset

What it does

This workflow applies CellChat to a single dataset to infer and visualize cell-cell communication networks. The committed materials use a human skin example and walk through object creation, database selection, interaction scoring, pathway aggregation, and downstream network visualization.

When to use it

Use this workflow when you already have normalized single-cell expression data and cell labels and want to identify likely ligand-receptor interactions between cell groups. It fits best as a downstream branch after clustering and annotation are already complete.

Prerequisites

Steps

Load normalized expression data and cell metadata

The example starts from an .rda object containing an expression matrix and metadata, then prepares inputs in the format CellChat expects.

load("./data/data_humanSkin_CellChat.rda")
cellchat <- createCellChat(object = data.input, meta = meta, group.by = "labels")

The notebook is written as a single-dataset workflow, so it expects normalized expression values and one grouping column that defines the cell labels used throughout the communication analysis.

Choose the interaction database and preprocess signaling genes

The committed example uses the human CellChat database and restricts the analysis to secreted signaling interactions before subsetting signaling genes and identifying overexpressed features.

CellChatDB <- CellChatDB.human
CellChatDB.use <- subsetDB(CellChatDB, search = "Secreted Signaling")
cellchat@DB <- CellChatDB.use

cellchat <- subsetData(cellchat)
cellchat <- identifyOverExpressedGenes(cellchat)

The source also documents an optional PPI projection step for shallow data, while noting that users can skip it and rely on raw.use = TRUE in computeCommunProb() if they want to avoid that smoothing stage.

Infer communication probabilities, filter sparse groups, and aggregate pathways

The central scoring step computes interaction probabilities, filters sparse groups, aggregates pathway-level networks, and prepares them for visualization.

cellchat <- computeCommunProb(cellchat)
cellchat <- filterCommunication(cellchat, min.cells = 10)
cellchat <- computeCommunProbPathway(cellchat)
cellchat <- aggregateNet(cellchat)

The notebook spends extra time explaining that the number of inferred interactions depends strongly on how average expression is computed, including a note about trimean versus truncated-mean settings.

Visualize pathway-level and ligand-receptor-level networks

After inference, the page branches into hierarchy, circle, chord, heatmap, and contribution plots for named signaling pathways and for individual ligand-receptor pairs.

netVisual_aggregate(cellchat, signaling = pathways.show, vertex.receiver = vertex.receiver)
netVisual_aggregate(cellchat, signaling = pathways.show, layout = "circle")
netAnalysis_contribution(cellchat, signaling = pathways.show)
netVisual_individual(cellchat, signaling = pathways.show, pairLR.use = LR.show, layout = "chord")

Compare specific sources and targets with bubble and chord views

The notebook then moves from whole-pathway summaries to selected source-target contrasts, using netVisual_bubble() and netVisual_chord_gene() to zoom in on specific communication patterns.

netVisual_bubble(cellchat, sources.use = 4, targets.use = c(5:11), remove.isolate = FALSE)
netVisual_chord_gene(cellchat, sources.use = c(1,2,3,4), targets.use = c(5:11), signaling = c("CCL", "CXCL"))

Run systems-level role and pattern analysis

The final section computes signaling centrality, sender/receiver roles, outgoing and incoming communication patterns, and structural/functional embeddings before saving the CellChat object.

cellchat <- netAnalysis_computeCentrality(cellchat, slot.name = "netP")
gg1 <- netAnalysis_signalingRole_scatter(cellchat)
ht1 <- netAnalysis_signalingRole_heatmap(cellchat, pattern = "outgoing")
netVisual_embedding(cellchat, type = "functional", label.size = 3.5)

Gotchas / notes

  • The committed example is adapted from a CellChat-style vignette and uses a bundled human skin dataset rather than a lab-specific object.
  • Database choice is species-specific; the notebook notes switching to CellChatDB.mouse for mouse data.
  • Interaction counts can change materially with the averaging method and filtering choices described in the notebook.
  • There are no committed standalone figure assets in this folder beyond the rendered HTML output.

📄 View source on GitHub