celltalker

The goal of celltalker is to infer putative ligand and receptor interactions from single-cell RNAseq data. This is accomplished by evaluating known cognate ligand/receptor interactions across groups of cells. Interactions are scored by jointly weighting the expression levels of ligands and receptors, and significance is evaluated by comparing to a background distribution of scrambled ligands and receptors.

A recent refactoring of this package has made the interface much simpler to use. We also provide a wrapper to the functionality of the circlize package for creating circos plots of ligands and receptors.

Installation

You can install the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("arc85/celltalker")

TL;DR example use

Here’s a basic use case for celltalker based on a subset of the 40,000 bone marrow single-cell dataset available from the Human Cell Atlas. Check out SeuratData here to download the example dataset.

Note that we have identified canonical immune cell types and filtered the dataset for ease of analysis. A vignette documenting this processing is coming soon, but please utilize the hca_bm_umap_cell_types data.frame for the vignette below.

# Load packages
suppressMessages({
  library(celltalker)
  library(Seurat)
  suppressWarnings(
    library(SeuratData)
  )
  library(tidyverse)
})

# Load Human Cell Atlast Bone Marrow from SeuratData
data(hcabm40k)

# Filter cell and assign cell types to the dataset
# NB: hca_bm_umap_cell_types has cell types and UMAP embeddings
hca_bm <- hcabm40k[,rownames(hca_bm_umap_cell_types)]
hca_bm[["cell_types"]] <- hca_bm_umap_cell_types$cell_types

# Process data
hca_bm <- NormalizeData(hca_bm)

# Add UMAP coordinates
hca_bm[["umap"]] <- CreateDimReducObject(embeddings=as.matrix(hca_bm_umap_cell_types[,1:2]),
  key="UMAP_",assay="RNA")

# View cell types
DimPlot(hca_bm,group.by="cell_types")

## Run celltalker
hca_bm_interactions <- celltalk(input_object=hca_bm,
  metadata_grouping="cell_types",
  ligand_receptor_pairs=ramilowski_pairs,
  number_cells_required=100,
  min_expression=1000,
  max_expression=20000,
  scramble_times=10)

## Identify top statistically significant interactions
top_stats <- hca_bm_interactions %>%
  mutate(fdr=p.adjust(p_val,method="fdr")) %>%
  filter(fdr<0.05) %>%
  group_by(cell_type1) %>%
  top_n(3,interact_ratio) %>%
  ungroup()

## Generate a circos plot
colors_use <- RColorBrewer::brewer.pal(n=length(unique(hca_bm$cell_types)),"Set2")

circos_plot(ligand_receptor_frame=top_stats,
  cell_group_colors=colors_use,
  ligand_color="blue",
  receptor_color="red",
  cex_outer=0.5,
  cex_inner=0.4)