Large-scale single-cell RNA-seq data manipulation with GDS files
Dr. Xiuwen Zheng (Genomics Research Center, AbbVie)
Dec 2020
Introduction
The SCArray package provides large-scale single-cell RNA-seq data manipulation using Genomic Data Structure (GDS) files. It combines dense/sparse matrices stored in GDS files and the Bioconductor infrastructure framework (SingleCellExperiment and DelayedArray) to provide out-of-memory data storage and manipulation using the R programming language. As shown in the figure, SCArray provides a SingleCellExperiment object for downstream data analyses. GDS is an alternative to HDF5. Unlike HDF5, GDS supports the direct storage of a sparse matrix without converting it to multiple vectors.
Figure 1: Workflow of SCArray
Installation
Requires R (>= v3.5.0), gdsfmt (>= v1.24.0)
Bioconductor repository
To install this package, start R and enter:
Format conversion
Conversion from SingleCellExperiment
The SCArray package can convert a single-cell experiment object (SingleCellExperiment) to a GDS file using the function scConvGDS(). For example,
suppressPackageStartupMessages(library(SCArray))
suppressPackageStartupMessages(library(SingleCellExperiment))
# load a SingleCellExperiment object
fn <- system.file("extdata", "LaMannoBrainSub.rds", package="SCArray")
sce <- readRDS(fn)
# convert to a GDS file
scConvGDS(sce, "test.gds")## Output: test.gds
## Compression: LZMA_RA
## Dimension: 18219 x 25
## Assay List:
## counts |+ counts { SparseReal32 18219x25 LZMA_ra(13.1%), 75.6K }
## rowData:
## colData:
## CELL_ID
## Cell_type
## Done.## Object of class "SCArrayFileClass"
## File: /tmp/RtmpiTt828/Rbuilde5b4343cb15b1/SCArray/vignettes/test.gds (137.9K)
## + [ ] *
## |--+ feature.id { Str8 18219 LZMA_ra(48.6%), 60.5K }
## |--+ sample.id { Str8 25 LZMA_ra(40.0%), 157B }
## |--+ counts { SparseReal32 18219x25 LZMA_ra(13.1%), 75.6K }
## |--+ feature.data [ ]
## |--+ sample.data [ ]
## | |--+ CELL_ID { Str8 25 LZMA_ra(40.0%), 157B }
## | \--+ Cell_type { Str8 25 LZMA_ra(64.3%), 133B }
## \--+ meta.data [ ]Conversion from a matrix
The input of scConvGDS() can be a dense or sparse matrix for count data:
library(Matrix)
cnt <- matrix(0, nrow=4, ncol=8)
set.seed(100); cnt[sample.int(length(cnt), 8)] <- rpois(8, 4)
(cnt <- as(cnt, "dgCMatrix"))## 4 x 8 sparse Matrix of class "dgCMatrix"
##
## [1,] . . . . . . . 6
## [2,] 3 1 . . . 4 . .
## [3,] . . . . . 3 . 4
## [4,] 4 . 3 . . . . .## Output: test.gds
## Compression: LZMA_RA
## Dimension: 4 x 8
## Assay List:
## counts |+ counts { SparseReal32 4x8 LZMA_ra(159.4%), 109B }
## Done.Single cell datasets
When a single-cell GDS file is available, users can use scExperiment() to load a SingleCellExperiment object from the GDS file. The assay data in the SingleCellExperiment object are DelayedMatrix objects to avoid the memory limit.
# a GDS file in the SCArray package
(fn <- system.file("extdata", "LaMannoBrainData.gds", package="SCArray"))## [1] "/tmp/RtmpiTt828/Rinste5b43493ea507/SCArray/extdata/LaMannoBrainData.gds"## class: SingleCellExperiment
## dim: 12000 243
## metadata(0):
## assays(1): counts
## rownames(12000): Rp1 Sox17 ... Efhd2 Fhad1
## rowData names(0):
## colnames(243): 1772072122_A04 1772072122_A05 ... 1772099011_H05
## 1772099012_E04
## colData names(2): CELL_ID Cell_type
## reducedDimNames(0):
## mainExpName: NULL
## altExpNames(0):## <12000 x 243> sparse matrix of class SC_GDSMatrix and type "double":
## 1772072122_A04 1772072122_A05 ... 1772099011_H05 1772099012_E04
## Rp1 0 0 . 0 0
## Sox17 0 0 . 0 0
## Mrpl15 1 2 . 2 2
## Lypla1 0 0 . 0 1
## Tcea1 1 0 . 6 1
## ... . . . . .
## Agmat 0 0 . 0 0
## Dnajc16 0 0 . 0 0
## Casp9 0 0 . 0 0
## Efhd2 0 0 . 1 1
## Fhad1 0 0 . 1 0## DataFrame with 243 rows and 2 columns
## CELL_ID Cell_type
## <character> <character>
## 1772072122_A04 1772072122_A04 DA-VTA4
## 1772072122_A05 1772072122_A05 DA-VTA2
## 1772072122_A06 1772072122_A06 DA-VTA3
## 1772072122_A07 1772072122_A07 DA-SNC
## 1772072122_A08 1772072122_A08 DA-VTA4
## ... ... ...
## 1772099011_D01 1772099011_D01 DA-VTA1
## 1772099011_F04 1772099011_F04 DA-VTA2
## 1772099011_G07 1772099011_G07 DA-VTA4
## 1772099011_H05 1772099011_H05 DA-SNC
## 1772099012_E04 1772099012_E04 DA-VTA4## DataFrame with 12000 rows and 0 columnsData Manipulation and Analysis
SCArray provides a SingleCellExperiment object for downstream data analyses. At first, we create a log count matrix logcnt from the count matrix. Note that logcnt is also a DelayedMatrix without actually generating the whole matrix.
## <12000 x 243> sparse matrix of class DelayedMatrix and type "double":
## 1772072122_A04 1772072122_A05 ... 1772099011_H05 1772099012_E04
## Rp1 0.000000 0.000000 . 0.000000 0.000000
## Sox17 0.000000 0.000000 . 0.000000 0.000000
## Mrpl15 1.000000 1.584963 . 1.584963 1.584963
## Lypla1 0.000000 0.000000 . 0.000000 1.000000
## Tcea1 1.000000 0.000000 . 2.807355 1.000000
## ... . . . . .
## Agmat 0 0 . 0 0
## Dnajc16 0 0 . 0 0
## Casp9 0 0 . 0 0
## Efhd2 0 0 . 1 1
## Fhad1 0 0 . 1 0Mean for each column or row
The DelayedMatrixStats package provides functions operating on rows and columns of DelayedMatrix objects. For example, we can calculate the mean for each column or row of the log count matrix.
suppressPackageStartupMessages(library(DelayedMatrixStats))
col_mean <- DelayedMatrixStats::colMeans2(logcnt)
str(col_mean)## num [1:243] 0.261 0.138 0.238 0.259 0.143 ...## num [1:12000] 0 0.00652 0.81827 0.47055 1.33912 ...UMAP analysis
The scater package can perform the uniform manifold approximation and projection (UMAP) for the cell data, based on the data in a SingleCellExperiment object.
plotReducedDim() plots cell-level reduced dimension results (UMAP) stored in the SingleCellExperiment object:
Session Info
## R version 4.2.0 RC (2022-04-19 r82224)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.4 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.15-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.15-bioc/R/lib/libRlapack.so
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_GB 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] stats4 stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] scater_1.24.0 ggplot2_3.3.5
## [3] scuttle_1.6.0 DelayedMatrixStats_1.18.0
## [5] SingleCellExperiment_1.18.0 SummarizedExperiment_1.26.0
## [7] Biobase_2.56.0 GenomicRanges_1.48.0
## [9] GenomeInfoDb_1.32.0 SCArray_1.4.0
## [11] DelayedArray_0.22.0 IRanges_2.30.0
## [13] S4Vectors_0.34.0 MatrixGenerics_1.8.0
## [15] matrixStats_0.62.0 BiocGenerics_0.42.0
## [17] Matrix_1.4-1 gdsfmt_1.32.0
##
## loaded via a namespace (and not attached):
## [1] viridis_0.6.2 sass_0.4.1 BiocSingular_1.12.0
## [4] viridisLite_0.4.0 jsonlite_1.8.0 bslib_0.3.1
## [7] assertthat_0.2.1 highr_0.9 GenomeInfoDbData_1.2.8
## [10] vipor_0.4.5 yaml_2.3.5 ggrepel_0.9.1
## [13] pillar_1.7.0 lattice_0.20-45 glue_1.6.2
## [16] beachmat_2.12.0 digest_0.6.29 XVector_0.36.0
## [19] colorspace_2.0-3 cowplot_1.1.1 htmltools_0.5.2
## [22] pkgconfig_2.0.3 zlibbioc_1.42.0 purrr_0.3.4
## [25] scales_1.2.0 RSpectra_0.16-1 ScaledMatrix_1.4.0
## [28] BiocParallel_1.30.0 tibble_3.1.6 farver_2.1.0
## [31] generics_0.1.2 ellipsis_0.3.2 withr_2.5.0
## [34] cli_3.3.0 magrittr_2.0.3 crayon_1.5.1
## [37] evaluate_0.15 fansi_1.0.3 FNN_1.1.3
## [40] beeswarm_0.4.0 tools_4.2.0 lifecycle_1.0.1
## [43] stringr_1.4.0 munsell_0.5.0 irlba_2.3.5
## [46] compiler_4.2.0 jquerylib_0.1.4 rsvd_1.0.5
## [49] rlang_1.0.2 grid_4.2.0 RCurl_1.98-1.6
## [52] BiocNeighbors_1.14.0 labeling_0.4.2 bitops_1.0-7
## [55] rmarkdown_2.14 gtable_0.3.0 DBI_1.1.2
## [58] R6_2.5.1 gridExtra_2.3 knitr_1.38
## [61] dplyr_1.0.8 uwot_0.1.11 fastmap_1.1.0
## [64] utf8_1.2.2 stringi_1.7.6 ggbeeswarm_0.6.0
## [67] parallel_4.2.0 Rcpp_1.0.8.3 vctrs_0.4.1
## [70] tidyselect_1.1.2 xfun_0.30 sparseMatrixStats_1.8.0