In drisso/scRNAseq: Collection of Public Single-Cell RNA-Seq Datasets

library(BiocStyle)
knitr::opts_chunk$set(error=FALSE, message=FALSE, warning=FALSE)

Downloading the data

We obtain a single-cell RNA sequencing dataset of human prefrontal cortex cells from @zhong2018singlecell. Counts for endogenous genes are available from the Gene Expression Omnibus using the accession number GSE104276. We download and cache them using the r Biocpkg("BiocFileCache") package.

library(BiocFileCache)
bfc <- BiocFileCache("raw_data", ask = FALSE)
fname <- bfcrpath(bfc, 
    "https://www.ncbi.nlm.nih.gov/geo/download/?acc=GSE104276&format=file&file=GSE104276%5Fall%5Fpfc%5F2394%5FUMI%5Fcount%5FNOERCC%2Exls%2Egz")
tmp <- tempfile(fileext=".xls")
R.utils::gunzip(fname, destname=tmp, remove=FALSE)

Reading all the counts in as sparse matrices. Despite its name, this is not actually an XLS file.

library(scuttle)
counts <- readSparseCounts(tmp, skip.row=1, col.names=FALSE)
colnames(counts) <- strsplit(readLines(tmp, n=1), "\t")[[1]]
dim(counts)

We pull down some sample-level metadata in SOFT format.

library(GEOquery)
out <- GEOquery::getGEO("GSE104276")

df <- as(phenoData(out[[1]]), "data.frame")
sampdata <- DataFrame(
    developmental_stage=df[["developmental stage:ch1"]],
    gender=df[["gender:ch1"]],
    sample=df$title
)

# Fixing an error in their annotation.
sampdata$developmental_stage[sampdata$developmental_stage == "week gestation"] <- "23 weeks after gestation"

sampdata

Unfortunately, it's not enough to link the exact samples to their cells. We take the nuclear option of rolling through the per-sample TPM files to figure out who lives where.

fname <- bfcrpath(bfc, 
    "https://www.ncbi.nlm.nih.gov/geo/download/?acc=GSE104276&format=file")

tmp <- tempfile()
untar(fname, exdir=tmp)

all.files <- list.files(tmp, full=TRUE)
out <- lapply(all.files, function(x) {
    strsplit(readLines(x, n=1), "\t")[[1]]
})

# So many errors in their sample names...
samples <- sub("\\..*", "", basename(all.files))
samples <- sub("GSM[0-9]+_", "", samples)
samples <- sub("GW8", "GW08", samples)
samples <- sub("GW9", "GW09", samples)
samples <- sub("GW19_PFC", "GW19_PFC1_", samples)

mapping <- rep(basename(samples), lengths(out))
names(mapping) <- unlist(out)

converted <- mapping[colnames(counts)]
stopifnot(all(!is.na(converted)))
m <- match(converted, sampdata$sample)
stopifnot(all(!is.na(m)))

coldata <- sampdata[m,]
rownames(coldata) <- colnames(counts)
coldata

But that's not all, because we can add even more annotation about the cell type.

fname <- bfcrpath(bfc, 
    "https://www.ncbi.nlm.nih.gov/geo/download/?acc=GSE104276&format=file&file=GSE104276%5Freadme%5Fsample%5Fbarcode%2Exlsx")

library(readxl)
more.data <- read_xlsx(fname, sheet="SampleInfo")
more.data <- as.data.frame(more.data)
head(more.data)

m <- match(rownames(coldata), more.data[,1])
coldata <- cbind(coldata, more.data[m,c(-1, -ncol(more.data))])
colnames(coldata)

Saving to file

We now save all of the relevant components to file for upload to r Biocpkg("ExperimentHub").

path <- file.path("scRNAseq", "zhong-prefrontal", "2.6.0")
dir.create(path, showWarnings=FALSE, recursive=TRUE)
saveRDS(counts, file=file.path(path, "counts.rds"))
saveRDS(coldata, file=file.path(path, "coldata.rds"))