Bioconductor Scdblfinder

scDblFinder

Dependencies & Environment

Package-intrinsic requirements from the Bioconductor landing page — reproduce in any R environment.

• Version: 1.26.0 · Bioconductor: 3.23 · R: ≥ 4.6
• Depends: SingleCellExperiment
• Imports: igraph, Matrix, BiocGenerics, BiocParallel, BiocNeighbors, BiocSingular, S4Vectors, SummarizedExperiment, scran, scater, scuttle, bluster, DelayedArray, xgboost, MASS, IRanges, GenomicRanges, GenomeInfoDb, Rsamtools, rtracklayer
• Install: BiocManager::install("scDblFinder")

When to Use

• Identifying heterotypic doublets in single-cell RNA sequencing data using the scDblFinder iterative classifier.
• Detecting doublets in multiplexed samples by processing each capture separately using the samples argument.
• Testing for specific doublet type enrichments (e.g., combinations of clusters) using doubletPairwiseEnrichment or clusterStickiness.
• Recovering intra-sample doublets that are neighbors to known inter-sample doublets (e.g., from cell hashing) using recoverDoublets.

When NOT to Use

• For identifying homotypic doublets or inter-sample doublets in multiplexed experiments (use cell hashing or genotype-based SNP calls instead, as scDblFinder focuses on transcriptionally distinct heterotypic doublets).

Data Requirements

• A SingleCellExperiment object or a simple count matrix containing raw counts (assay 'counts').
• Empty droplets should already be removed prior to running the tool.

Key Parameters

• samples (NULL): Column name in colData or a vector indicating the sample/capture of origin to process batches separately.
• clusters (FALSE): Cluster labels or TRUE to use fastcluster; enables cluster-based artificial doublet generation.
• dbr (NULL): Expected proportion of doublets; if omitted, automatically estimated based on the number of captured cells.
• dbr.per1k (0.008): Expected doublet rate per thousand cells, defaulting to standard 10X rates.
• artificialDoublets (NULL): Number of artificial doublets to generate; defaults to roughly the number of cells.
• nfeatures (1000): Number of top expressed genes to retain for dimensionality reduction to speed up analysis.

Best Practices

• Always split multi-sample datasets by providing the samples argument so doublets are sought independently for each capture.
• Use the cluster-based approach (clusters=TRUE) if your dataset has a very clear cluster structure to avoid generating unidentifiable homotypic artificial doublets.
• Check the distribution of doublet scores (hist(sce$scDblFinder.score)); a bimodal distribution indicates successful doublet identification.
• If running multiple samples with the cluster-based approach, pre-cluster all samples together using fastcluster and pass the labels to the clusters argument to ensure concordant labels across samples.

Common Pitfalls

• Way too many doublets called: Often caused by pooling multiple samples without specifying the samples argument, making the tool assume a massive single capture. Fix: Provide the samples argument to split cells by capture/batch.
• 'Size factors should be positive' error: Caused by cells with zero reads or very low read counts that drop to zero after feature selection. Fix: Filter out empty droplets and extremely low-count cells before running scDblFinder.
• Nonsensical cluster labels across samples: When running multiple samples without providing cluster labels, scDblFinder clusters them sample-wise, leading to mismatched labels. Fix: Pre-cluster all samples together using fastcluster and pass the labels to the clusters argument.

Alternatives

• computeDoubletDensity: A simpler density-based method (formerly scran::doubletCells) that identifies cells with a high local density of artificial doublets.
• findDoubletClusters: Identifies entire clusters that are likely composed of doublets by checking if their expression profile lies between two other clusters.
• directDblClassification: Trains a classifier directly on gene expression without the kNN step, though generally yields worse predictions than scDblFinder.

Citations

• Germain, P. L., Lun, A., Macnair, W., & Robinson, M. D. (2021). Doublet identification in single-cell sequencing data using scDblFinder. F1000Research, 10, 979.
• Howitt et al., 2024. (Referenced regarding doublets in 10x Flex datasets).

References

• Homepage: bioconductor.org/packages/scDblFinder
• Vignette: vignette_0_d20f7c0c.txt

Run this on BioMate

This skill is the knowledge layer — when, why, and how to use scdblfinder. To run this analysis on your own data with managed compute, automated QC, and reproducible outputs, use BioMate — free to start.

▶ Open scdblfinder on BioMate →