bio-single-cell-doublet-detection

name	bio-single-cell-doublet-detection
description	Detect and remove doublets (multiple cells captured in one droplet) from single-cell RNA-seq data. Uses Scrublet (Python), DoubletFinder (R), and scDblFinder (R). Essential QC step before clustering to avoid artificial cell populations. Use when identifying and removing doublets from scRNA-seq data.
tool_type	mixed
primary_tool	Scrublet

Doublet Detection

Doublets are droplets containing two or more cells. They appear as artificial intermediate cell populations and must be removed before analysis.

Scrublet (Python)

Fast doublet detection based on simulated doublets from the data.

Basic Usage

import scrublet as scr
import scanpy as sc
import numpy as np

adata = sc.read_10x_mtx('filtered_feature_bc_matrix/')

scrub = scr.Scrublet(adata.X, expected_doublet_rate=0.06)
doublet_scores, predicted_doublets = scrub.scrub_doublets()

adata.obs['doublet_score'] = doublet_scores
adata.obs['predicted_doublet'] = predicted_doublets

print(f'Detected {predicted_doublets.sum()} doublets ({100*predicted_doublets.mean():.1f}%)')

Adjust Parameters

scrub = scr.Scrublet(adata.X, expected_doublet_rate=0.06)
doublet_scores, predicted_doublets = scrub.scrub_doublets(
    min_counts=2,
    min_cells=3,
    min_gene_variability_pctl=85,
    n_prin_comps=30,
    synthetic_doublet_umi_subsampling=1.0
)

Visualize Doublet Scores

import matplotlib.pyplot as plt

scrub.plot_histogram()
plt.savefig('doublet_histogram.pdf')

# UMAP with doublet scores
sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata)
sc.pp.pca(adata)
sc.pp.neighbors(adata)
sc.tl.umap(adata)

sc.pl.umap(adata, color=['doublet_score', 'predicted_doublet'], save='_doublets.pdf')

Filter Doublets

adata_filtered = adata[~adata.obs['predicted_doublet']].copy()
print(f'Kept {adata_filtered.n_obs} cells after doublet removal')

Set Manual Threshold

scrub = scr.Scrublet(adata.X)
doublet_scores, _ = scrub.scrub_doublets()

threshold = 0.25
predicted_doublets = doublet_scores > threshold
adata.obs['predicted_doublet'] = predicted_doublets

DoubletFinder (R)

Popular R package for doublet detection in Seurat workflows.

Basic Usage

library(Seurat)
library(DoubletFinder)

seurat_obj <- Read10X(data.dir = 'filtered_feature_bc_matrix/')
seurat_obj <- CreateSeuratObject(counts = seurat_obj, min.cells = 3, min.features = 200)

seurat_obj <- NormalizeData(seurat_obj)
seurat_obj <- FindVariableFeatures(seurat_obj)
seurat_obj <- ScaleData(seurat_obj)
seurat_obj <- RunPCA(seurat_obj)
seurat_obj <- RunUMAP(seurat_obj, dims = 1:20)
seurat_obj <- FindNeighbors(seurat_obj, dims = 1:20)
seurat_obj <- FindClusters(seurat_obj, resolution = 0.5)

sweep.res  paramSweepseurat_obj PCs   sct  
sweep.stats  summarizeSweepsweep.res GT  
bcmvn  find.pKsweep.stats

optimal_pk  bcmvnpKwhich.maxbcmvnBCmetric

nExp_poi    nrowseurat_objmeta.data
seurat_obj  doubletFinderseurat_obj PCs   pN   pK  optimal_pk
                             nExp  nExp_poi reuse.pANN   sct  

colnamesseurat_objmeta.data

With SCTransform

seurat_obj <- SCTransform(seurat_obj)
seurat_obj <- RunPCA(seurat_obj)
seurat_obj <- RunUMAP(seurat_obj, dims = 1:30)
seurat_obj <- FindNeighbors(seurat_obj, dims = 1:30)
seurat_obj <- FindClusters(seurat_obj, resolution = 0.5)

sweep.res <- paramSweep(seurat_obj, PCs = 1:30, sct = TRUE)
sweep.stats <- summarizeSweep(sweep.res, GT = FALSE)
bcmvn <- find.pK(sweep.stats)

optimal_pk <- as.numeric(as.character(bcmvn$pK[which.maxbcmvnBCmetric
nExp_poi    nrowseurat_objmeta.data

seurat_obj  doubletFinderseurat_obj PCs   pN   pK  optimal_pk
                             nExp  nExp_poi reuse.pANN   sct  

Filter Doublets

df_col <- grep('DF.classifications', colnames([e-mail]), value = TRUE)
seurat_obj$doublet <- [e-mail][[df_col]]

DimPlot(seurat_obj, group.by = 'doublet')

seurat_obj <- subset(seurat_obj, subset = doublet == 'Singlet')

Adjust Expected Doublet Rate

n_cells <- ncol(seurat_obj)
doublet_rate <- n_cells / 1000 * 0.008
nExp_poi <- round(doublet_rate * n_cells)

scDblFinder (R/Bioconductor)

Fast Bioconductor package using gradient boosting for doublet detection.

Basic Usage

library(scDblFinder)
library(SingleCellExperiment)

sce <- SingleCellExperiment(assays = list(counts = counts_matrix))
sce <- scDblFinder(sce)

table(sce$scDblFinder.class)

From Seurat Object

library(scDblFinder)
library(Seurat)

sce <- as.SingleCellExperiment(seurat_obj)

sce <- scDblFinder(sce)

seurat_obj$scDblFinder_class <- sce$scDblFinder.class
seurat_obj$scDblFinder_score <- sce$scDblFinder.score

DimPlot(seurat_obj, group.by = 'scDblFinder_class')

seurat_obj <- subset(seurat_obj, subset = scDblFinder_class == 'singlet')

Multi-Sample Processing

sce <- scDblFinder(sce, samples = 'sample_id')

Adjust Parameters

sce <- scDblFinder(sce,
    dbr = 0.06,
    dbr.sd = 0.015,
    nfeatures = 1500,
    dims = 20,
    k = 30
)

Expected Doublet Rates

Cells Loaded	Expected Rate
1,000	~0.8%
2,000	~1.6%
5,000	~4.0%
10,000	~8.0%
15,000	~12%

Formula: rate ≈ cells_loaded / 1000 * 0.008

Compare Methods

library(scDblFinder)

seurat_obj$scrublet <- scrublet_results
sce <- as.SingleCellExperiment(seurat_obj)
sce <- scDblFinder(sce)
seurat_obj$scDblFinder <- sce$scDblFinder.class

DimPlot(seurat_obj, group.by = c('doublet', 'scDblFinder', 'scrublet'), ncol = 3)

table(seurat_obj$doublet, seurat_obj$scDblFinder)

Handling Heterotypic vs Homotypic Doublets

Heterotypic Doublets

• Two different cell types
• Easier to detect (intermediate expression)
• All methods handle well

Homotypic Doublets

• Same cell type
• Harder to detect (no intermediate signature)
• May have higher total counts

adata.obs['log_counts'] = np.log1p(adata.obs['total_counts'])
sc.pl.violin(adata, 'log_counts', groupby='predicted_doublet')

Scanpy Integration Pipeline

import scanpy as sc
import scrublet as scr

adata = sc.read_10x_mtx('filtered_feature_bc_matrix/')

adata.var['mt'] = adata.var_names.str.startswith('MT-')
sc.pp.calculate_qc_metrics(adata, qc_vars=['mt'], inplace=True)

scrub = scr.Scrublet(adata.X, expected_doublet_rate=0.06)
doublet_scores, predicted_doublets = scrub.scrub_doublets()
adata.obs['doublet_score'] = doublet_scores
adata.obs['is_doublet'] = predicted_doublets

print(f'Before filtering: {adata.n_obs} cells')
adata = adata[~adata.obs['is_doublet']].copy()
adata = adata[adata.obs['pct_counts_mt'] < 20].copy()
print(f'After filtering: {adata.n_obs} cells')

sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata)
sc.pp.pca(adata)
sc.pp.neighbors(adata)
sc.tl.umap(adata)
sc.tl.leiden(adata)

Seurat Integration Pipeline

library(Seurat)
library(DoubletFinder)

seurat_obj <- Read10X('filtered_feature_bc_matrix/')
seurat_obj <- CreateSeuratObject(counts = seurat_obj, min.cells = 3, min.features = 200)

seurat_obj[['percent.mt']] <- PercentageFeatureSet(seurat_obj, pattern = '^MT-')

seurat_obj <- NormalizeData(seurat_obj)
seurat_obj <- FindVariableFeatures(seurat_obj)
seurat_obj <- ScaleData(seurat_obj)
seurat_obj <- RunPCA(seurat_obj)
seurat_obj <- RunUMAP(seurat_obj, dims = 1:20)
seurat_obj <- FindNeighbors(seurat_obj, dims = 1
seurat_obj  FindClustersseurat_obj resolution  

sweep.res  paramSweepseurat_obj PCs  
sweep.stats  summarizeSweepsweep.res
bcmvn  find.pKsweep.stats
pk  bcmvnpKwhich.maxbcmvnBCmetric
nExp    ncolseurat_obj

seurat_obj  doubletFinderseurat_obj PCs   pN   pK  pk nExp  nExp

df_col  grep colnamesseurat_objmeta.data value  
seurat_obj  subsetseurat_obj cells  colnamesseurat_objseurat_objmeta.datadf_col  
seurat_obj  subsetseurat_obj subset  percent.mt  

seurat_obj  NormalizeDataseurat_obj
seurat_obj  FindVariableFeaturesseurat_obj
seurat_obj  ScaleDataseurat_obj
seurat_obj  RunPCAseurat_obj
seurat_obj  RunUMAPseurat_obj dims  
seurat_obj  FindNeighborsseurat_obj dims  
seurat_obj  FindClustersseurat_obj

Method Comparison

Method	Speed	Accuracy	Language
Scrublet	Fast	Good	Python
DoubletFinder	Slow	Good	R
scDblFinder	Fast	Excellent	R

Related Skills

• preprocessing - QC before doublet detection
• clustering - Run after filtering doublets
• data-io - Load data before processing