Nebulosa Gene Expression Visualization

Kernel density estimation for visualizing gene expression across single cells, recovering dropout signal and revealing co-expression patterns.

Overview

Nebulosa is an R package that visualizes single-cell data using kernel density estimation (KDE). Standard scatter plots of gene expression can be misleading in single-cell data due to dropout — genes that are expressed but recorded as zero. Nebulosa recovers this signal by incorporating cell-to-cell similarity in its density estimates, producing smooth, interpretable expression maps.

Key advantages over raw FeaturePlot():

• Recovers signal from dropped-out features by pooling information from similar cells
• Removes spurious “random” expression in areas not supported by many cells
• Enables joint density visualization to identify co-expressing cell populations

Citation: Jose Alquicira-Hernandez and Joseph E. Powell. Nebulosa recovers single cell gene expression signals by kernel density estimation. doi: 10.18129/B9.bioc.Nebulosa

Source: powellgenomicslab/Nebulosa

Installation

remotes::install_github('powellgenomicslab/Nebulosa')

Key function

plot_density() is the main function from the Nebulosa package. Its interface resembles Seurat’s FeaturePlot(), making it easy to drop into existing Seurat workflows.

Complete workflow

1. Load libraries

   library(Nebulosa)
   library(Seurat)
   library(BiocFileCache)
   
2. Load and preprocess data

   This example uses a 3k PBMC dataset from 10x Genomics:

   bfc <- BiocFileCache(ask = FALSE)
   data_file <- bfcrpath(bfc,
     file.path("https://s3-us-west-2.amazonaws.com/10x.files/samples/cell",
               "pbmc3k", "pbmc3k_filtered_gene_bc_matrices.tar.gz")
   )
   untar(data_file, exdir = tempdir())
   
   data <- Read10X(
     data.dir = file.path(tempdir(), "filtered_gene_bc_matrices", "hg19")
   )
   
   pbmc <- CreateSeuratObject(
     counts = data,
     project = "pbmc3k",
     min.cells = 3,
     min.features = 200
   )
   
3. Filter low-quality cells

   pbmc[["percent.mt"]] <- PercentageFeatureSet(pbmc, pattern = "^MT-")
   pbmc <- subset(pbmc, subset = nFeature_RNA < 2500 & percent.mt < 5)
   
4. Normalize and reduce dimensions

   Nebulosa works on any 2D embedding. Here, use SCTransform followed by PCA and UMAP:

   pbmc <- SCTransform(pbmc, verbose = FALSE)
   pbmc <- RunPCA(pbmc)
   pbmc <- RunUMAP(pbmc, dims = 1:30)
   
5. Cluster

   pbmc <- FindNeighbors(pbmc, dims = 1:30)
   pbmc <- FindClusters(pbmc)
   
6. Visualize with Nebulosa

   Plot the kernel density estimate for a single gene:

   plot_density(pbmc, "CD4")
   

   Compare with Seurat’s standard feature plot:

   # Standard Seurat plot
   FeaturePlot(pbmc, "CD4")
   FeaturePlot(pbmc, "CD4", order = TRUE)
   
   # Nebulosa density plot — smoother, recovers dropout
   plot_density(pbmc, "CD4")
   

   Nebulosa removes the “random” scattered expression of CD4 in areas where it is not biologically supported, while still highlighting CD4+ T cells and myeloid cells.

Multi-feature visualization

Nebulosa supports plotting multiple features simultaneously and computing joint densities to identify co-expressing populations.

Individual densities for multiple genes

# Returns individual density plots for each gene
p <- plot_density(pbmc, c("CD8A", "CCR7"))
p + plot_layout(ncol = 1)

Joint density

Use joint = TRUE to multiply the per-gene densities into a single joint density plot. This highlights cells that co-express all queried genes:

# Joint density of CD8A and CCR7 — identifies Naive CD8+ T cells
p <- plot_density(pbmc, c("CD8A", "CCR7"), joint = TRUE)
p + plot_layout(ncol = 1)

Accessing individual plots

Set combine = FALSE to get a list of ggplot objects. The last element is always the joint density:

p_list <- plot_density(pbmc, c("CD8A", "CCR7"), joint = TRUE, combine = FALSE)

# Individual gene densities
p_list[[1]]  # CD8A
p_list[[2]]  # CCR7

# Joint density
p_list[[length(p_list)]]

Identifying cell populations with joint density

# Naive CD4+ T cells: CD4 and CCR7 co-expression
p4 <- plot_density(pbmc, c("CD4", "CCR7"), joint = TRUE)
p4 + plot_layout(ncol = 1)

# Compare joint density against cluster labels
p4[[3]] / DimPlot(pbmc, label = TRUE, repel = TRUE)

When to use Nebulosa

Nebulosa is most valuable for:

• Dropped-out genes — genes with high dropout rates where raw expression plots are sparse and hard to interpret
• Co-expression analysis — identifying populations that express multiple markers simultaneously
• Communication in presentations — smoother density maps are often clearer for figures and talks

For genes with strong, high-coverage expression, standard FeaturePlot() may be equally informative. Use Nebulosa alongside core Seurat visualization methods to draw more informed conclusions.

Additional resources

• Nebulosa GitHub repository