Overview of Doublet Detecting Softwares

Transcription-based doublet detection softwares use the transcriptomic profiles in each cell to predict whether that cell is a singlet or doublet. Most methods simulate doublets by adding the transcriptional profiles of two droplets in your pool together. Therefore, these approaches assume that only a small percentage of the droplets in your dataset are doublets. The table bellow provides a comparison of the different methods.

Doublet Detecting Software

QC Filtering Required

Requires Pre-clustering

Doublet Detecting Method

DoubletDecon

✖️

✔️

Deconvolution based on clusters provided.

DoubletDetection

✖️

✖️

Iterative boost classifier to classify doublets.

DoubletFinder

✔️

✖️

Identify ideal cluster size and call expected number of droplets with highest number of simulated doublet neighbors as doublets.

scDblFinder

✖️

✖️

Gradient boosted trees trained with number neighboring doublets and QC metrics to classify doublets

Scds

✖️

✖️

cxds: Uses genes pairs that are typically not expressed in the same droplet to rank droplets based on co-expression of all pairs.
bcds: Uses highly variable genes and simulated doublets to train a binary classification algorithm and return probability of droplet being a doublet.

Scrublet

✖️

✖️

Identifies the number of neighboring simulated doublets for each droplet and uses bimodal distribution of scores to classify singlets and doublets.

Solo

✖️

✖️

Simulates doublets and fits a two-layer neural network.

If you don’t know which demultiplexing software(s) to run, take a look at our Software Selection Recommendations based on your dataset.