H5weaver

Read, Rearrange, and Write HDF5 files with 10x Genomics conventions

Requirements

H5weaver requires libraries for HDF5 files. On Windows, these are bundled with the rhdf5 package. On Linux/Unix, you will need to first install hdf5 libraries.

This can usually be accomplished with:

sudo apt-get install hdf5-dev

or

sudo yum install hdf5-dev

Once hdf5 libraries are available, you can proceed to installation of rhdf5.

The rhdf5 package is provided through BioConductor, and can be installed using:

if(!"BiocManager" %in% .packages(all.available = TRUE)) {
  install.packages("BiocManager")
}
BiocManager::install("rhdf5")

H5weaver also requires the data.table, ids, and Matrix packages, which are available on CRAN and should be automatically installed by install_github().

Installation

This package can be installed from Github using the devtools package.

You may first need to register your GitHub PAT, as this is a private repository.

Get access token from github:

• Navigate to Settings / Developer settings
• Click Personal access tokens
• Generate new token (or re-generate if you have an existing one but you didn't copy it to your password manager 😒).
• Under Select scopes
• Give the token repo scope

from github:

Make sure to copy your new personal access token now. You won’t be able to see it again!

Sys.setenv(GITHUB_PAT = "your-access-token-here")
devtools::install_github("aifimmunology/H5weaver")

AIFI .h5 structure

The structure of AIFI .h5 files differs somewhat from files supplied directly by CellRanger. The major difference is the addition of richer metadata and additional results (e.g. from Cell Hashing).

A summary of the structure is available in Google Sheets here: 10x .h5 file structure

Reading .h5 files

H5weaver allows for reading of the contents of an HDF5 file in multiple ways. To demonstrate, we'll use a file stored in the H5weaver package. You can obtain the path to this file using:

h5_file <- system.file("testdata/well1.h5", package = "H5weaver")

Reading as a Seurat object

We can read and .h5 file directoy into a Seurat object using read_h5_seurat():

library(H5weaver)

so <- read_h5_seurat(h5_file)

This function places the RNA-seq counts in the "RNA" assay, and CITE-seq ADT counts (if present) in the "ADT" assay.

Reading as a SingleCellExperiment object

Likewise, we can read directly into a SingleCellExperiment object for use with BioConductor packages using read_h5_sce().

library(H5weaver)

sce <- read_h5_sce(h5_file)

Note that this requires a recent version of SingleCellExperiment (>= 1.8.0) so that ADTs are handled correctly.

In this case, RNA-$eq counts are stored in the "counts" slot for the SingleCellExperiment, and CITE-seq ADT counts (if present) are stored in the altExp "ADT" slot.

Reading the matrix directly

There is a convenience function to directly read the main cell x gene matrix from the HDF5 file, read_h5_dgCMatrix():

library(H5weaver)

mat <- read_h5_dgCMatrix(h5_file)

Note: By default, this matrix will be 1-indexed for convenient use in R. If you would rather retrieve a 0-indexed matrix, set the index1 parameter to FALSE:

mat <- read_h5_dgCMatrix(h5_file,
                         index1 = FALSE)

Reading cell metadata directly

A convenience function is provided to retrieve all cell/observation-based metadata, read_h5_cell_meta():

cell_meta <- read_h5_cell_meta(h5_file)

Note that for the test dataset, this is only the cell barcodes, as additional metadata are not present.

Reading feature metadata directly

A similar function is also provided for gene/feature-based metadata, read_h5_feature_meta():

feat_meta <- read_h5_feature_meta(h5_file)

Reading and separating out all contents

To read the entirety of an HDF5 file as a list object, use h5dump():

library(H5weaver)

h5_list <- h5dump(h5_file)
str(h5_list)

This is a very raw representation of the contents of these HDF5 files. You may want to convert the major components to a sparse matrix (for cell x gene counts), and a data.frame (for metadata):

h5_list <- h5_list_convert_to_dgCMatrix(h5_list,
                                        target = "matrix")
                                        
mat <- h5_list$matrix_dgCMatrix

feature_metadata <- as.data.frame(h5_list$matrix$features[-1])

Now, mat will consist of a dgCMatrix with genes as rows and barcodes as columns, and feature_metadata will be a data.frame with genes as rows and various metadata as columns.

For this test dataset, there isn't any cell metadata. However, files that are generated by our pipeline will include a substantial metadata set stored in matrix/observations. This can be retrieved with:

cell_metadata <- cbind(data.frame(barcodes = h5_list$matrix$barcodes),
                       as.data.frame(h5_list$matrix$observations))

Tests

Tests for HTOparser are implemented using the testthat testing suite:
https://testthat.r-lib.org/

To run tests for this package, download the repository from Github and run devtools::test() in R from the base directory of the package.

Extra-stringent, CRAN-level package testing can be performed using devtools::check() in R.

Style and Standards

This package aims to conform to the tidyverse style guide:
https://style.tidyverse.org/index.html

General information about R package conventions can be found in R Packages:
http://r-pkgs.had.co.nz/