Skip to contents

Doublet removal

Source: vignettes/doublet_removal.Rmd
doublet_removal.Rmd

First set up Zorn/Bascet according to the install instructions. This tutorial assumes that you have already produced a Seurat object using one of the count workflows, e.g. the KRAKEN2 workflow, the alignment-based workflow, the informative KMER workflow, or the count sketch workflow. You should have a count table of some sorts ready for analysis.

When to remove doublets

A doublet in single-cell metagenomics is a barcode that captures DNA from more than one cell. These can arise for at least two common reasons:

  1. If more cells are collected than there are barcodes, you get barcode collisions. Droplet and split-pool methods rely on the cells being much fewer than the avaiable barcodes. By doing “Poisson loading” (diluting enough), doublets are typically kept <2%; but they still occur and must be handled

  2. If cells stick together, the will also receive the same barcode. Ensuring a good single-cell suspension is crucial

Zorn/Bascet produces count tables and you can thus use many single-cell tools to process the data, such as to remove doublets. Here we give an example using scDblFinder.

Removing doublets using scDblFinder

scDblFinder lives in Bioconductor and operates on a SingleCellExperiment object, so the workflow is to wrap the count matrix from your Seurat object, run the caller, then copy the per-cell scores back into the Seurat object’s metadata.

library(SingleCellExperiment)
library(scDblFinder)

## Wrap the raw counts as a SingleCellExperiment (rows = features, cols = cells)
sce <- SingleCellExperiment::SingleCellExperiment(
  list(counts = GetAssayData(adata,layer = "counts"))
)

## Call doublets
sce <- scDblFinder::scDblFinder(sce)

## Copy the per-cell score and class back into the Seurat object
adata$scDblFinder.score <- sce$scDblFinder.score
adata$scDblFinder.class <- sce$scDblFinder.class

You can now overlay the score and class on the UMAP to see whether doublets form a coherent population (often they sit between two species clusters, or they all form a “blob” in the middle of the UMAP):

FeaturePlot(adata, features = "scDblFinder.score")
DimPlot(adata, group.by = "scDblFinder.class")

To remove doublets, just subset the Seurat object:

adata <- adata[, adata$scDblFinder.class == "singlet"]