1 Class structure

1.1 Introduction

The SpatialExperiment class is an R/Bioconductor S4 class for storing data from spatially resolved transcriptomics (ST) experiments. The class extends the SingleCellExperiment class for single-cell data to support storage and retrieval of additional information from spot-based and molecule-based ST platforms, including spatial coordinates, images, and image metadata. A specialized constructor function is included for data from the 10x Genomics Visium platform.

The following schematic illustrates the SpatialExperiment class structure.

Overview of the SpatialExperiment class structure.

Figure 1: Overview of the SpatialExperiment class structure

As shown, an object consists of: (i) assays containing expression counts, (ii) rowData containing information on features, i.e. genes, (iii) colData containing information on spots or cells, including nonspatial and spatial metadata, (iv) spatialCoords containing spatial coordinates, and (v) imgData containing image data. For spot-based ST data (e.g. 10x Genomics Visium), a single assay named counts is used. For molecule-based ST data (e.g. seqFISH), two assays named counts and molecules can be used.

Additional details on the class structure are provided in our preprint.

1.2 Load data

For demonstration of the general class structure, we load an example SpatialExperiment (abbreviated as SPE) object (variable spe):

library(SpatialExperiment)
example(read10xVisium, echo = FALSE)
spe
## class: SpatialExperiment 
## dim: 50 99 
## metadata(0):
## assays(1): counts
## rownames(50): ENSMUSG00000051951 ENSMUSG00000089699 ...
##   ENSMUSG00000005886 ENSMUSG00000101476
## rowData names(1): symbol
## colnames(99): AAACAACGAATAGTTC-1 AAACAAGTATCTCCCA-1 ...
##   AAAGTCGACCCTCAGT-1 AAAGTGCCATCAATTA-1
## colData names(4): in_tissue array_row array_col sample_id
## reducedDimNames(0):
## mainExpName: NULL
## altExpNames(0):
## spatialCoords names(2) : pxl_col_in_fullres pxl_row_in_fullres
## imgData names(4): sample_id image_id data scaleFactor

1.3 spatialCoords

In addition to observation metadata stored inside the colData slot, the SpatialExperiment class stores spatial coordinates as:

  • spatialCoords, a numeric matrix of spatial coordinates (e.g. x and y)

spatialCoords are stored inside the int_colData, and are directly accessible via the corresponding accessor:

head(spatialCoords(spe))
##                    pxl_col_in_fullres pxl_row_in_fullres
## AAACAACGAATAGTTC-1               2312               1252
## AAACAAGTATCTCCCA-1               8230               7237
## AAACAATCTACTAGCA-1               4170               1611
## AAACACCAATAACTGC-1               2519               8315
## AAACAGAGCGACTCCT-1               7679               2927
## AAACAGCTTTCAGAAG-1               1831               6400

The corresponding column names can be also accessed with spatialCoordsNames():

spatialCoordsNames(spe)
## [1] "pxl_col_in_fullres" "pxl_row_in_fullres"

1.4 imgData

All image related data are stored inside the int_metadata’s imgData field as a DataFrame of the following structure:

  • each row corresponds to one image for a given sample
    and with a given unique image identifier (e.g. its resolutions)
  • for each image, columns specify:
    • which sample_id the image belongs to
    • a unique image_id in order to accommodate multiple images
      for a given sample (e.g. of different resolutions)
    • the image’s data (a SpatialImage object)
    • the scaleFactor that adjusts pixel positions of the original,
      full-resolution image to pixel positions in the image

The imgData() accessor can be used to retrieve the image data stored within the object:

imgData(spe)
## DataFrame with 2 rows and 4 columns
##     sample_id    image_id   data scaleFactor
##   <character> <character> <list>   <numeric>
## 1    section1      lowres   ####   0.0510334
## 2    section2      lowres   ####   0.0510334

1.4.1 The SpatialImage class

Images are stored inside the data field of the imgData as a list of SpatialImages. Each image may be of one of the following sub-classes:

  • LoadedSpatialImage
    • represents an image that is fully realized into memory as a raster object
    • @image contains a raster object: a matrix of RGB colors for each pixel in the image
  • StoredSpatialImage
    • represents an image that is stored in a local file (e.g., as
      .png, .jpg or .tif), and loaded into memory only on request
    • @path specifies a local file from which to retrieve the image
  • RemoteSpatialImage
    • represents an image that is remotely hosted
      (under some URL), and retrieved only on request
    • @url specifies where to retrieve the image from

A SpatialImage can be accessed using getImg(), or retrieved directly from the imgData():

(spi <- getImg(spe))
## 576 x 600 (width x height) StoredSpatialImage
## imgSource(): 
##   /tmp/RtmpecNr2O/Rinst1be6bc7e8463dd/SpatialExperiment/extdata/10xVisium/section1
##   /outs/spatial/tissue_lowres_image.png
identical(spi, imgData(spe)$data[[1]])
## [1] TRUE

Data available in an object of class SpatialImage may be accessed via the imgRaster() and imgSource() accessors:

plot(imgRaster(spe))

1.4.2 Adding or removing images

Image entries may be added or removed from a SpatialExperiment’s imgData DataFrame using addImg() and rmvImg(), respectively.

Besides a path or URL to source the image from and a numeric scale factor, addImg() requires specification of the sample_id the new image belongs to, and an image_id that is not yet in use for that sample:

url <- "https://i.redd.it/3pw5uah7xo041.jpg"
spe <- addImg(spe, 
    sample_id = "section1", 
    image_id = "pomeranian",
    imageSource = url, 
    scaleFactor = NA_real_, 
    load = TRUE)
img <- imgRaster(spe, 
    sample_id = "section1", 
    image_id = "pomeranian")
plot(img)

The rmvImg() function is more flexible in the specification of the sample_id and image_id arguments. Specifically:

  • TRUE is equivalent to all, e.g.
    sample_id = "<sample>", image_id = TRUE
    will drop all images for a given sample
  • NULL defaults to the first entry available, e.g.
    sample_id = "<sample>", image_id = NULL
    will drop the first image for a given sample

For example, sample_id = TRUE, image_id = TRUE will specify all images; sample_id = NULL, image_id = NULL corresponds to the first image entry in the imgData; sample_id = TRUE, image_id = NULL equals the first image for all samples; and sample_id = NULL, image_id = TRUE matches all images for the first sample.

Here, we remove section1’s pomeranian image added in the previous code chunk; the image is now completely gone from the imgData:

imgData(spe <- rmvImg(spe, "section1", "pomeranian"))
## DataFrame with 2 rows and 4 columns
##     sample_id    image_id   data scaleFactor
##   <character> <character> <list>   <numeric>
## 1    section1      lowres   ####   0.0510334
## 2    section2      lowres   ####   0.0510334

2 Object construction

2.1 Manually

The SpatialExperiment constructor provides several arguments to give maximum flexibility to the user.

In particular, these include:

  • spatialCoords, a numeric matrix containing spatial coordinates
  • spatialCoordsNames, a character vector specifying which
    colData fields correspond to spatial coordinates

spatialCoords can be supplied via colData by specifying the column names that correspond to spatial coordinates with spatialCoordsNames:

n <- length(z <- letters)
y <- matrix(nrow = n, ncol = n)
cd <- DataFrame(x = seq(n), y = seq(n), z)

spe1 <- SpatialExperiment(
    assay = y, 
    colData = cd, 
    spatialCoordsNames = c("x", "y"))

Alternatively, spatialCoords may be supplied separately as a matrix, e.g.:

xy <- as.matrix(cd[, c("x", "y")])

spe2 <- SpatialExperiment(
    assay = y, 
    colData = cd["z"], 
    spatialCoords = xy)

Importantly, both of the above SpatialExperiment() function calls lead to construction of the exact same object:

identical(spe1, spe2)
## [1] TRUE

Finally, spatialCoords(Names) are optional, i.e., we can construct a SPE using only a subset of the above arguments:

spe <- SpatialExperiment(
    assays = y)
isEmpty(spatialCoords(spe))
## [1] TRUE

In general, spatialCoordsNames takes precedence over spatialCoords, i.e., if both are supplied, the latter will be ignored. In other words, spatialCoords are preferentially extracted from the DataFrame provided via colData. E.g., in the following function call, spatialCoords will be ignored, and xy-coordinates are instead extracted from the input colData according to the specified spatialCoordsNames. In this case, a message is also provided:

n <- 10; m <- 20
y <- matrix(nrow = n, ncol = m)
cd <- DataFrame(x = seq(m), y = seq(m))
xy <- matrix(nrow = m, ncol = 2)
colnames(xy) <- c("x", "y")

SpatialExperiment(
    assay = y, 
    colData = cd,
    spatialCoordsNames = c("x", "y"),
    spatialCoords = xy)
## Both 'spatialCoords' and 'spatialCoordsNames'  have been supplied;
## using 'spatialCoords'. Set either to NULL to suppress this message.

2.2 Spot-based

When working with spot-based ST data, such as 10x Genomics Visium or other platforms providing images, it is possible to store the image information in the dedicated imgData structure.

Also, the SpatialExperiment class stores a sample_id value in the colData structure, which is possible to set with the sample_id argument (default is “sample_01”).

Here we show how to load the default Space Ranger data files from a 10x Genomics Visium experiment, and build a SpatialExperiment object.

In particular, the readImgData() function is used to build an imgData DataFrame to be passed to the SpatialExperiment constructor. The sample_id used to build the imgData object must be the same one used to build the SpatialExperiment object, otherwise an error is returned.

dir <- system.file(
   file.path("extdata", "10xVisium", "section1", "outs"),
   package = "SpatialExperiment")

# read in counts
fnm <- file.path(dir, "raw_feature_bc_matrix")
sce <- DropletUtils::read10xCounts(fnm)

# read in image data
img <- readImgData(
    path = file.path(dir, "spatial"),
    sample_id = "foo")

# read in spatial coordinates
fnm <- file.path(dir, "spatial", "tissue_positions_list.csv")
xyz <- read.csv(fnm, header = FALSE,
    col.names = c(
        "barcode", "in_tissue", "array_row", "array_col",
        "pxl_row_in_fullres", "pxl_col_in_fullres"))

# construct observation & feature metadata
rd <- S4Vectors::DataFrame(
    symbol = rowData(sce)$Symbol)

# construct 'SpatialExperiment'
(spe <- SpatialExperiment(
    assays = list(counts = assay(sce)),
    rowData = rd, 
    colData = DataFrame(xyz), 
    spatialCoordsNames = c("pxl_col_in_fullres", "pxl_row_in_fullres"),
    imgData = img,
    sample_id = "foo"))
## class: SpatialExperiment 
## dim: 50 50 
## metadata(0):
## assays(1): counts
## rownames(50): ENSMUSG00000051951 ENSMUSG00000089699 ...
##   ENSMUSG00000005886 ENSMUSG00000101476
## rowData names(1): symbol
## colnames: NULL
## colData names(5): barcode in_tissue array_row array_col sample_id
## reducedDimNames(0):
## mainExpName: NULL
## altExpNames(0):
## spatialCoords names(2) : pxl_col_in_fullres pxl_row_in_fullres
## imgData names(4): sample_id image_id data scaleFactor

Alternatively, the read10xVisium() function facilitates the import of 10x Genomics Visium data to handle one or more samples organized in folders reflecting the default Space Ranger folder tree organization, as illustrated below (where “raw/filtered” refers to either “raw” or “filtered” to match the data argument). Note that the base directory “outs/” from Space Ranger can either be included manually in the paths provided in the samples argument, or can be ignored; if ignored, it will be added automatically. The .h5 files are used if type = "HDF5". (Note that tissue_positions.csv was renamed in Space Ranger v2.0.0.)

sample
 . | — outs
 · · | — raw/filtered_feature_bc_matrix.h5
 · · | — raw/filtered_feature_bc_matrix
 · · · | — barcodes.tsv.gz
 · · · | — features.tsv.gz
 · · · | — matrix.mtx.gz
 · · | — spatial
 · · · | — scalefactors_json.json
 · · · | — tissue_lowres_image.png
 · · · | — tissue_positions.csv

Using read10xVisium(), the above code to construct the same SPE is reduced to:

dir <- system.file(
    file.path("extdata", "10xVisium"),
    package = "SpatialExperiment")

sample_ids <- c("section1", "section2")
samples <- file.path(dir, sample_ids, "outs")

(spe10x <- read10xVisium(samples, sample_ids,
    type = "sparse", data = "raw",
    images = "lowres", load = FALSE))
## class: SpatialExperiment 
## dim: 50 99 
## metadata(0):
## assays(1): counts
## rownames(50): ENSMUSG00000051951 ENSMUSG00000089699 ...
##   ENSMUSG00000005886 ENSMUSG00000101476
## rowData names(1): symbol
## colnames(99): AAACAACGAATAGTTC-1 AAACAAGTATCTCCCA-1 ...
##   AAAGTCGACCCTCAGT-1 AAAGTGCCATCAATTA-1
## colData names(4): in_tissue array_row array_col sample_id
## reducedDimNames(0):
## mainExpName: NULL
## altExpNames(0):
## spatialCoords names(2) : pxl_col_in_fullres pxl_row_in_fullres
## imgData names(4): sample_id image_id data scaleFactor

Or alternatively, omitting the base directory outs/ from Space Ranger:

samples2 <- file.path(dir, sample_ids)

(spe10x2 <- read10xVisium(samples2, sample_ids,
    type = "sparse", data = "raw",
    images = "lowres", load = FALSE))
## class: SpatialExperiment 
## dim: 50 99 
## metadata(0):
## assays(1): counts
## rownames(50): ENSMUSG00000051951 ENSMUSG00000089699 ...
##   ENSMUSG00000005886 ENSMUSG00000101476
## rowData names(1): symbol
## colnames(99): AAACAACGAATAGTTC-1 AAACAAGTATCTCCCA-1 ...
##   AAAGTCGACCCTCAGT-1 AAAGTGCCATCAATTA-1
## colData names(4): in_tissue array_row array_col sample_id
## reducedDimNames(0):
## mainExpName: NULL
## altExpNames(0):
## spatialCoords names(2) : pxl_col_in_fullres pxl_row_in_fullres
## imgData names(4): sample_id image_id data scaleFactor

2.3 Molecule-based

To demonstrate how to accommodate molecule-based ST data (e.g. seqFISH platform) inside a SpatialExperiment object, we generate some mock data of 1000 molecule coordinates across 50 genes and 20 cells. These should be formatted into a data.frame where each row corresponds to a molecule, and columns specify the xy-positions as well as which gene/cell the molecule has been assigned to:

n <- 1e3  # number of molecules
ng <- 50  # number of genes
nc <- 20  # number of cells
# sample xy-coordinates in [0, 1]
x <- runif(n)
y <- runif(n)
# assign each molecule to some gene-cell pair
gs <- paste0("gene", seq(ng))
cs <- paste0("cell", seq(nc))
gene <- sample(gs, n, TRUE)
cell <- sample(cs, n, TRUE)
# assure gene & cell are factors so that
# missing observations aren't dropped
gene <- factor(gene, gs)
cell <- factor(cell, cs)
# construct data.frame of molecule coordinates
df <- data.frame(gene, cell, x, y)
head(df)
##     gene   cell          x         y
## 1  gene1  cell1 0.55265761 0.6567272
## 2 gene49  cell4 0.90808261 0.1133595
## 3 gene48 cell12 0.56241271 0.1883192
## 4 gene50 cell12 0.71094524 0.1895197
## 5 gene24  cell2 0.47053029 0.9140180
## 6 gene32 cell17 0.05129459 0.3819849

Next, it is possible to re-shape the above table into a BumpyMatrix using splitAsBumpyMatrix(), which takes as input the xy-coordinates, as well as arguments specifying the row and column index of each observation:

# construct 'BumpyMatrix'
library(BumpyMatrix)
mol <- splitAsBumpyMatrix(
    df[, c("x", "y")], 
    row = gene, col = cell)

Finally, it is possible to construct a SpatialExperiment object with two data slots:

  • The counts assay stores the number of molecules per gene and cell
    (equivalent to transcript counts in spot-based data)
  • The molecules assay holds the spatial molecule positions (xy-coordinates)

Each entry in the molecules assay is a DFrame that contains the positions of all molecules from a given gene that have been assigned to a given cell.

# get count matrix
y <- with(df, table(gene, cell))
y <- as.matrix(unclass(y))
y[1:5, 1:5]
##        cell
## gene    cell1 cell2 cell3 cell4 cell5
##   gene1     2     4     1     1     0
##   gene2     1     2     4     0     0
##   gene3     0     1     1     2     0
##   gene4     0     1     2     0     2
##   gene5     0     1     1     2     0
# construct SpatialExperiment
spe <- SpatialExperiment(
    assays = list(
        counts = y, 
        molecules = mol))
spe
## class: SpatialExperiment 
## dim: 50 20 
## metadata(0):
## assays(2): counts molecules
## rownames(50): gene1 gene2 ... gene49 gene50
## rowData names(0):
## colnames(20): cell1 cell2 ... cell19 cell20
## colData names(1): sample_id
## reducedDimNames(0):
## mainExpName: NULL
## altExpNames(0):
## spatialCoords names(0) :
## imgData names(0):

The BumpyMatrix of molecule locations can be accessed using the dedicated molecules() accessor:

molecules(spe)
## 50 x 20 BumpyDataFrameMatrix
## rownames: gene1 gene2 ... gene49 gene50 
## colnames: cell1 cell2 ... cell19 cell20 
## preview [1,1]:
##   DataFrame with 2 rows and 2 columns
##             x         y
##     <numeric> <numeric>
##   1  0.552658 0.6567272
##   2  0.862979 0.0847017

3 Common operations

3.1 Subsetting

Subsetting objects is automatically defined to synchronize across all attributes, as for any other Bioconductor Experiment class.

For example, it is possible to subset by sample_id as follows:

sub <- spe10x[, spe10x$sample_id == "section1"]

Or to retain only observations that map to tissue via:

sub <- spe10x[, colData(spe10x)$in_tissue]
sum(colData(spe10x)$in_tissue) == ncol(sub)
## [1] TRUE

3.2 Combining samples

To work with multiple samples, the SpatialExperiment class provides the cbind method, which assumes unique sample_id(s) are provided for each sample.

In case the sample_id(s) are duplicated across multiple samples, the cbind method takes care of this by appending indices to create unique sample identifiers.

spe1 <- spe2 <- spe
spe3 <- cbind(spe1, spe2)
## 'sample_id's are duplicated across 'SpatialExperiment' objects to cbind; appending sample indices.
unique(spe3$sample_id)
## [1] "sample01.1" "sample01.2"

Alternatively (and preferentially), we can create unique sample_id(s) prior to cbinding as follows:

# make sample identifiers unique
spe1 <- spe2 <- spe
spe1$sample_id <- paste(spe1$sample_id, "A", sep = ".")
spe2$sample_id <- paste(spe2$sample_id, "B", sep = ".")

# combine into single object
spe3 <- cbind(spe1, spe2)

3.3 Sample ID replacement

In particular, when trying to replace the sample_id(s) of a SpatialExperiment object, these must map uniquely with the already existing ones, otherwise an error is returned.

new <- spe3$sample_id
new[1] <- "section2.A"
spe3$sample_id <- new
## Error in .local(x, ..., value): Number of unique 'sample_id's is 2, but 3 were provided.
new[1] <- "third.one.of.two"
spe3$sample_id <- new
## Error in .local(x, ..., value): Number of unique 'sample_id's is 2, but 3 were provided.

Importantly, the sample_id colData field is protected, i.e., it will be retained upon attempted removal (= replacement by NULL):

# backup original sample IDs
tmp <- spe$sample_id
# try to remove sample IDs
spe$sample_id <- NULL
# sample IDs remain unchanged
identical(tmp, spe$sample_id)
## [1] TRUE

3.4 Image transformations

Both the SpatialImage (SpI) and SpatialExperiment (SpE) class currently support two basic image transformations, namely, rotation (via rotateImg()) and mirroring (via mirrorImg()). Specifically, for a SpI/E x:

  • rotateImg(x, degrees) expects as degrees a single numeric in +/-[0,90,…,360].
    Here, a (negative) positive value corresponds to (counter-)clockwise rotation.
  • mirrorImg(x, axis) expects as axis a character string specifying
    whether to mirror horizontally ("h") or vertically ("v").

Here, we apply various transformations using both a SpI (spi) and SpE (spe) as input, and compare the resulting images to the original:

3.4.1 Rotation

# extract first image
spi <- getImg(spe10x)
# apply counter-/clockwise rotation
spi1 <- rotateImg(spi, -90)
spi2 <- rotateImg(spi, +90)
# visual comparison
par(mfrow = c(1, 3))
plot(as.raster(spi))
plot(as.raster(spi1))
plot(as.raster(spi2))

# specify sample & image identifier
sid <- "section1"
iid <- "lowres"
# counter-clockwise rotation
tmp <- rotateImg(spe10x, 
    sample_id = sid, 
    image_id = iid,
    degrees = -90)
# visual comparison
par(mfrow = c(1, 2))
plot(imgRaster(spe10x, sid, iid))
plot(imgRaster(tmp, sid, iid))

3.4.2 Mirroring

# extract first image
spi <- getImg(spe10x)
# mirror horizontally/vertically
spi1 <- mirrorImg(spi, "h")
spi2 <- mirrorImg(spi, "v")
# visual comparison
par(mfrow = c(1, 3))
plot(as.raster(spi))
plot(as.raster(spi1))
plot(as.raster(spi2))

# specify sample & image identifier
sid <- "section2"
iid <- "lowres"
# mirror horizontally
tmp <- mirrorImg(spe10x, 
    sample_id = sid, 
    image_id = iid,
    axis = "h")
# visual comparison
par(mfrow = c(1, 2))
plot(imgRaster(spe10x, sid, iid))
plot(imgRaster(tmp, sid, iid))

4 Session Info

sessionInfo()
## R version 4.2.2 (2022-10-31)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.5 LTS
## 
## Matrix products: default
## BLAS:   /home/biocbuild/bbs-3.16-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.16-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] BumpyMatrix_1.6.0           SpatialExperiment_1.8.1    
##  [3] SingleCellExperiment_1.20.0 SummarizedExperiment_1.28.0
##  [5] Biobase_2.58.0              GenomicRanges_1.50.2       
##  [7] GenomeInfoDb_1.34.9         IRanges_2.32.0             
##  [9] S4Vectors_0.36.2            BiocGenerics_0.44.0        
## [11] MatrixGenerics_1.10.0       matrixStats_0.63.0         
## [13] BiocStyle_2.26.0           
## 
## loaded via a namespace (and not attached):
##  [1] locfit_1.5-9.7            xfun_0.37                
##  [3] bslib_0.4.2               beachmat_2.14.0          
##  [5] HDF5Array_1.26.0          lattice_0.20-45          
##  [7] rhdf5_2.42.0              htmltools_0.5.4          
##  [9] yaml_2.3.7                rlang_1.0.6              
## [11] R.oo_1.25.0               jquerylib_0.1.4          
## [13] scuttle_1.8.4             R.utils_2.12.2           
## [15] BiocParallel_1.32.5       dqrng_0.3.0              
## [17] GenomeInfoDbData_1.2.9    zlibbioc_1.44.0          
## [19] R.methodsS3_1.8.2         codetools_0.2-19         
## [21] evaluate_0.20             knitr_1.42               
## [23] fastmap_1.1.1             curl_5.0.0               
## [25] parallel_4.2.2            highr_0.10               
## [27] Rcpp_1.0.10               edgeR_3.40.2             
## [29] formatR_1.14              BiocManager_1.30.20      
## [31] cachem_1.0.7              limma_3.54.2             
## [33] DelayedArray_0.24.0       magick_2.7.3             
## [35] jsonlite_1.8.4            XVector_0.38.0           
## [37] rjson_0.2.21              digest_0.6.31            
## [39] bookdown_0.32             grid_4.2.2               
## [41] cli_3.6.0                 tools_4.2.2              
## [43] bitops_1.0-7              rhdf5filters_1.10.0      
## [45] magrittr_2.0.3            sass_0.4.5               
## [47] RCurl_1.98-1.10           Matrix_1.5-3             
## [49] DelayedMatrixStats_1.20.0 sparseMatrixStats_1.10.0 
## [51] rmarkdown_2.20            Rhdf5lib_1.20.0          
## [53] R6_2.5.1                  DropletUtils_1.18.1      
## [55] compiler_4.2.2