Read Cell Ranger HDF5 .h5 files in R

read_cellranger.R

# install.packages(c("Matrix", "rhdf5", "tidyverse"))
library(Matrix)
library(rhdf5)
library(tidyverse)
library(glue)

my_h5_files <- Sys.glob(
  "path/to/cellranger-per-channel/output/*/filtered_feature_bc_matrix.h5"
)

my_counts_list <- lapply(my_h5_files, function(my_h5_file) {
  # Each cell gets a barcode nucleotide sequence.
  # e.g. "AAACCTGAGACCGGAT-1"
  my_barcodes <- as.character(h5read(my_h5_file, "matrix/barcodes"))
  # We'll assume that the name of the folder containing the .h5 file
  # is the name of the channel (sample).
  my_channel <- basename(dirname(my_h5_file))
  # Prepend the sample name to the barcodes, so we can safely
  # merge multiple samples into a single matrix.
  my_barcodes <- sprintf("%s|%s", my_channel, my_barcodes)
  # Read the data into a dgCMatrix (sparse matrix).
  h5 <- h5read(my_h5_file, "matrix")
  counts <- sparseMatrix(
    dims = h5$shape,
    i = as.numeric(h5$indices),
    p = as.numeric(h5$indptr),
    x = as.numeric(h5$data),
    index1 = FALSE
  )
  colnames(counts) <- my_barcodes
  rownames(counts) <- as.data.frame(h5[["features"]])$id
  return(counts)
})

# Make sure all the count matrices have the same dimensions
stopifnot(length(unique(lapply(my_counts_list, dim))) == 1)

# Concatenate the matrices into one big matrix.
counts <- do.call(cbind, my_counts_list)
rm(my_counts_list)

# Compute the sparsity of the matrix.
sparsity <- 1 - length(counts@x) / (as.numeric(nrow(counts)) * ncol(counts))
n_genes <- nrow(counts)
n_cells <- ncol(counts)
print(glue("Counts matrix has {comma(n_genes)} genes and {comma(n_cells)} cells"))
print(glue("Counts matrix is {signif(100 * sparsity, 3)}% sparse"))

# We'll also want to get the corresponding gene information from the .h5 file.
genes <- h5read(my_h5_files[1], "matrix/features")
genes <- tibble(ensembl_id = genes$id, symbol = genes$name)