scds:single cell doublet scoring: In-silico doublet annotation for single cell RNA sequencing data
27 October 2020
Contents
1 Introduction
In this vignette, we provide an overview of the basic functionality and usage of the scds package, which interfaces with SingleCellExperiment objects.
2 Installation
Install the scds package using Bioconductor:
if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("scds", version = "3.9")Or from github:
library(devtools)
devtools::install_github('kostkalab/scds')3 Quick start
scds takes as input a SingleCellExperiment object (see here SingleCellExperiment), where raw counts are stored in a counts assay, i.e. assay(sce,"counts"). An example dataset created by sub-sampling the cell-hashing cell-lines data set (see https://satijalab.org/seurat/hashing_vignette.html) is included with the package and accessible via data("sce").Note that scds is designed to workd with larger datasets, but for the purposes of this vignette, we work with a smaller example dataset. We apply scds to this data and compare/visualize reasults:
3.1 Example data set
Get example data set provided with the package.
library(scds)
library(scater)
library(rsvd)
library(Rtsne)
library(cowplot)
set.seed(30519)
data("sce_chcl")
sce = sce_chcl #- less typing
dim(sce)## [1] 2000 2000We see it contains 2,000 genes and 2,000 cells, 216 of which are identified as doublets:
table(sce$hto_classification_global)##
## Doublet Negative Singlet
## 216 83 1701We can visualize cells/doublets after projecting into two dimensions:
logcounts(sce) = log1p(counts(sce))
vrs = apply(logcounts(sce),1,var)
pc = rpca(t(logcounts(sce)[order(vrs,decreasing=TRUE)[1:100],]))
ts = Rtsne(pc$x[,1:10],verb=FALSE)
reducedDim(sce,"tsne") = ts$Y; rm(ts,vrs,pc)
plotReducedDim(sce,"tsne",col="hto_classification_global")
3.2 Computational doublet annotation
We now run the scds doublet annotation approaches. Briefly, we identify doublets in two complementary ways: cxds is based on co-expression of gene pairs and works with absence/presence calls only, while bcds uses the full count information and a binary classification approach using artificially generated doublets. cxds_bcds_hybrid combines both approaches, for more details please consult (this manuscript). Each of the three methods returns a doublet score, with higher scores indicating more “doublet-like” barcodes.
#- Annotate doublet using co-expression based doublet scoring:
sce = cxds(sce,retRes = TRUE)
sce = bcds(sce,retRes = TRUE,verb=TRUE)
sce = cxds_bcds_hybrid(sce)
par(mfcol=c(1,3))
boxplot(sce$cxds_score ~ sce$doublet_true_labels, main="cxds")
boxplot(sce$bcds_score ~ sce$doublet_true_labels, main="bcds")
boxplot(sce$hybrid_score ~ sce$doublet_true_labels, main="hybrid")
3.3 Visualizing gene pairs
For cxds we can identify and visualize gene pairs driving doublet annoataions, with the expectation that the two genes in a pair might mark different types of cells (see manuscript). In the following we look at the top three pairs, each gene pair is a row in the plot below:
scds =
top3 = metadata(sce)$cxds$topPairs[1:3,]
rs = rownames(sce)
hb = rowData(sce)$cxds_hvg_bool
ho = rowData(sce)$cxds_hvg_ordr[hb]
hgs = rs[ho]
l1 = ggdraw() + draw_text("Pair 1", x = 0.5, y = 0.5)
p1 = plotReducedDim(sce,"tsne",col=hgs[top3[1,1]])
p2 = plotReducedDim(sce,"tsne",col=hgs[top3[1,2]])
l2 = ggdraw() + draw_text("Pair 2", x = 0.5, y = 0.5)
p3 = plotReducedDim(sce,"tsne",col=hgs[top3[2,1]])
p4 = plotReducedDim(sce,"tsne",col=hgs[top3[2,2]])
l3 = ggdraw() + draw_text("Pair 3", x = 0.5, y = 0.5)
p5 = plotReducedDim(sce,"tsne",col=hgs[top3[3,1]])
p6 = plotReducedDim(sce,"tsne",col=hgs[top3[3,2]])
plot_grid(l1,p1,p2,l2,p3,p4,l3,p5,p6,ncol=3, rel_widths = c(1,2,2))4 Session Info
sessionInfo()## R version 4.0.3 (2020-10-10)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 18.04.5 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.12-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.12-bioc/R/lib/libRlapack.so
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=C
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] parallel stats4 stats graphics grDevices utils datasets
## [8] methods base
##
## other attached packages:
## [1] cowplot_1.1.0 Rtsne_0.15
## [3] rsvd_1.0.3 scater_1.18.0
## [5] ggplot2_3.3.2 SingleCellExperiment_1.12.0
## [7] SummarizedExperiment_1.20.0 Biobase_2.50.0
## [9] GenomicRanges_1.42.0 GenomeInfoDb_1.26.0
## [11] IRanges_2.24.0 S4Vectors_0.28.0
## [13] BiocGenerics_0.36.0 MatrixGenerics_1.2.0
## [15] matrixStats_0.57.0 scds_1.6.0
## [17] BiocStyle_2.18.0
##
## loaded via a namespace (and not attached):
## [1] Rcpp_1.0.5 lattice_0.20-41
## [3] digest_0.6.27 R6_2.4.1
## [5] plyr_1.8.6 evaluate_0.14
## [7] pillar_1.4.6 sparseMatrixStats_1.2.0
## [9] zlibbioc_1.36.0 rlang_0.4.8
## [11] data.table_1.13.2 irlba_2.3.3
## [13] magick_2.5.0 Matrix_1.2-18
## [15] rmarkdown_2.5 labeling_0.4.2
## [17] BiocNeighbors_1.8.0 BiocParallel_1.24.0
## [19] stringr_1.4.0 RCurl_1.98-1.2
## [21] munsell_0.5.0 beachmat_2.6.0
## [23] DelayedArray_0.16.0 vipor_0.4.5
## [25] compiler_4.0.3 BiocSingular_1.6.0
## [27] xfun_0.18 pkgconfig_2.0.3
## [29] ggbeeswarm_0.6.0 htmltools_0.5.0
## [31] tidyselect_1.1.0 gridExtra_2.3
## [33] tibble_3.0.4 GenomeInfoDbData_1.2.4
## [35] bookdown_0.21 viridisLite_0.3.0
## [37] crayon_1.3.4 dplyr_1.0.2
## [39] withr_2.3.0 bitops_1.0-6
## [41] grid_4.0.3 gtable_0.3.0
## [43] lifecycle_0.2.0 magrittr_1.5
## [45] pROC_1.16.2 scales_1.1.1
## [47] stringi_1.5.3 scuttle_1.0.0
## [49] farver_2.0.3 XVector_0.30.0
## [51] viridis_0.5.1 DelayedMatrixStats_1.12.0
## [53] ellipsis_0.3.1 generics_0.0.2
## [55] vctrs_0.3.4 xgboost_1.2.0.1
## [57] tools_4.0.3 beeswarm_0.2.3
## [59] glue_1.4.2 purrr_0.3.4
## [61] yaml_2.2.1 colorspace_1.4-1
## [63] BiocManager_1.30.10 knitr_1.30