In kasperdanielhansen/TENxPBMCData: PBMC data from 10X Genomics

require(knitr)
opts_chunk$set(error=FALSE, message=FALSE, warning=FALSE)

BiocStyle::markdown()

Introduction

The r Biocpkg("TENxPBMCData") package provides a R / Bioconductor resource for representing and manipulating nine different single-cell RNA-seq (scRNA-seq) and CITE-seq data sets on peripheral blood mononuclear cells (PBMC) generated by 10X Genomics:

1. pbmc68k
2. frozen_pbmc_donor_a
3. frozen_pbmc_donor_b
4. frozen_pbmc_donor_c
5. pbmc33k
6. pbmc3k
7. pbmc6k
8. pbmc4k
9. pbmc8k
10. pbmc5k-CITEseq

The number in the dataset title is roughly the number of cells in the experiment.

This package makes extensive use of the r Biocpkg("HDF5Array") package to avoid loading the entire data set in memory, instead storing the counts on disk as a HDF5 file and loading subsets of the data into memory upon request.

Note: The purpose of this package is to provide testing and example data for Bioconductor packages. We have done no processing of the "filtered" 10X scRNA-RNA or CITE-seq data; it is delivered as is.

Work flow

Loading the data

We use the TENxPBMCData function to download the relevant files from Bioconductor's ExperimentHub web resource. This includes the HDF5 file containing the counts, as well as the metadata on the rows (genes) and columns (cells). The output is a single SingleCellExperiment object from the r Biocpkg("SingleCellExperiment") package. This is equivalent to a SummarizedExperiment class but with a number of features specific to single-cell data.

library(TENxPBMCData)
tenx_pbmc4k <- TENxPBMCData(dataset = "pbmc4k")
tenx_pbmc4k

Note: of particular interest to some users might be the pbmc68k dataset for its size.

The first call to TENxPBMCData() may take some time due to the need to download some moderately large files. The files are then stored locally such that ensuing calls in the same or new sessions are fast. Use the dataset argument to select which dataset to download; values are visible through the function definition:

args(TENxPBMCData)

The count matrix itself is represented as a DelayedMatrix from the r Biocpkg("DelayedArray") package. This wraps the underlying HDF5 file in a container that can be manipulated in R. Each count represents the number of unique molecular identifiers (UMIs) assigned to a particular gene in a particular cell.

counts(tenx_pbmc4k)

Exploring the data

To quickly explore the data set, we compute some summary statistics on the count matrix. We tell the r Biocpkg("DelayedArray") block size to indicate that we can use up to 1 GB of memory for loading the data into memory from disk.

options(DelayedArray.block.size=1e9)

We are interested in library sizes colSums(counts(tenx_pbmc4k)), number of genes expressed per cell colSums(counts(tenx_pbmc4k) != 0), and average expression across cells rowMeans(counts(tenx_pbmc4k)). A naive implement might be

lib.sizes <- colSums(counts(tenx_pbmc4k))
n.exprs <- colSums(counts(tenx_pbmc4k) != 0L)
ave.exprs <- rowMeans(counts(tenx_pbmc4k))

More advanced analysis procedures are implemented in various Bioconductor packages - see the SingleCell biocViews for more details.

Saving computations

Saving the tenx_pbmc4k object in a standard manner, e.g.,

destination <- tempfile()
saveRDS(tenx_pbmc4k, file = destination)

saves the row-, column-, and meta-data as an R object, and remembers the location and subset of the HDF5 file from which the object is derived. The object can be read into a new R session with readRDS(destination), provided the HDF5 file remains in it's original location.

CITE-seq datasets

For CITE-seq datasets, both the transcriptomics data and the antibody capture data are available from a single SingleCellExperiment object. While the transcriptomics data can be accessed directly as described above, the antibody capture data should be accessed with the altExp function. Again, the resulting count matrix is represented as a DelayedMatrix.

tenx_pbmc5k_CITEseq <- TENxPBMCData(dataset = "pbmc5k-CITEseq")

counts(altExp(tenx_pbmc5k_CITEseq))

Session information

sessionInfo()

kasperdanielhansen/TENxPBMCData documentation built on June 20, 2021, 5:13 p.m.