DoubletFinder: Doublet detection in single-cell RNA sequencing data using artificial nearest neighbors

Abstract

Single-cell RNA sequencing (scRNA-seq) using droplet microfluidics occasionally 11 produces transcriptome data representing more than one cell. These technical artifacts are 12 caused by cell doublets formed during cell capture and occur at a frequency proportional to the 13 total number of sequenced cells. The presence of doublets can lead to spurious biological 14 conclusions, which justifies the practice of sequencing fewer cells to limit doublet formation rates. 15Here, we present a computational doublet detection tool -DoubletFinder -that identifies 16 doublets based solely on gene expression features. DoubletFinder infers the putative gene 17 expression profile of real doublets by generating artificial doublets from existing scRNA-seq data. 18Neighborhood detection in gene expression space then identifies sequenced cells with 19 increased probability of being doublets based on their proximity to artificial doublets. 20 DoubletFinder robustly identifies doublets across scRNA-seq datasets with variable numbers of 21 cells and sequencing depth, and predicts false-negative and false-positive doublets defined 22 using conventional barcoding approaches. We anticipate that DoubletFinder will aid in scRNA-23 seq data analysis and will increase the throughput and accuracy of scRNA-seq experiments. 24 25 INTRODUCTION 26Since its introduction nearly a decade ago, scRNA-seq has been used to elucidate 27 previously unknown cell types and reconstruct developmental dynamics among heterogeneous 28 cell populations (Human Cell Atlas Consortium, 2017). At first, scRNA-seq workflows were 29All rights reserved. No reuse allowed without permission.(which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.The copyright holder for this preprint . http://dx.doi.org/10.1101/352484 doi: bioRxiv preprint first posted online Jun. 20, 2018; 2 limited to tens to hundreds of cells which hindered data interpretation due to batch effects and 30 low statistical power (Stegle et al., 2016). Today, sequencing thousands to hundreds of 31 thousands of cells is routine due to the advent of droplet microfluidics and nanowell-based 32 sequencing strategies (Macosko et al., 2015; Klein et al., 2015; Zheng et al., 2017; Gierahn et 33 al., 2017; Takara Bio USA, 2018). These techniques rely on a Poisson loading strategy to 34 compartmentalize individual cells and mRNA capture beads before cell lysis, mRNA capture, 35 and transcript barcoding via reverse transcription. Since cells are captured randomly, the 36 proportion of droplets containing >1 cell -known as doublets -scales linearly across an 37 experimentally-relevant range of input cell concentrations (10X Genomics, 2017), justifying the 38 practice of limiting the number of sequenced cells to minimize doublet formation rates. 39The confounding effects of doublets in scRNA-seq data are well-appreciated (Ilicic et al., 40 2016). However, genomic and cellular barcoding techniques for identifying doublets have only 41 recently ...

Molecular identification of sixty-three amacrine cell types completes a mouse retinal cell atlas

“…Criteria for doublets included having an increased number of transcripts per cell and having combined expression of canonical markers from different cell classes. 10) We retested for doublets among amacrine cells using the R package 'DoubletFinder' (McGinnis, et al 2019) with the default parameter of 7.5% as expected doublet rate. We found that ~55% of cells in clusters 16, and 60 could be doublets, although we have no way of verifying this possibility.…”

Linked optical and gene expression profiling of single cells at high throughput

“…Moreover, PDM allows systematic variation of nanowell contents across the array, to choose conditions that maximize data quality (22). For example, controlled cell loading minimizes doublets, nanowells in which two cells are inadvertently sequenced as one, which can be major confounders in scRNA-seq experiments that assume single-cell data (25,26). Moreover, controlled printing allows us to load multiple capture beads to every well ( Fig.…”

Inference and effects of barcode multiplets in droplet-based single-cell assays

“…We envision a combination of dense exogenous barcodes via cell hashing 10 and evolved by CRISPR-Cas9 11 or intrinsic features such as clonal mutations, rearrangements, or highly correlated abundances with barcode sequence similarity metrics could be leveraged to better infer barcode multiplets. Such approaches would complement existing tools that robustly identify cell doublets 12,13 and empty droplets 14 from droplet-based scRNA-seq and further mitigate hidden confounders in single-cell data. Until then, we suggest that inferences regarding rare cell events should be corroborated across multiple channels or technologies to validate interpretation.…”