scran.chan is a slimmed-down version of the scran Bioconductor package, refactored to use the libscran C++ library for all of the computation. This provides methods for an end-to-end analysis of a single-cell RNA-sequencing (scRNA-seq) analysis, starting from the count matrix and finishes with clusters, markers, and various embeddings (i.e., t-SNE and UMAP). It's pretty fast and memory-efficient - some casual timings show that 170,000 cells can be analyzed in 5 minutes with 8 threads.
Compilation is straightforward provided you have a reasonably up-to-date CMake and a compiler that supports C++17.
devtools::install_github("LTLA/scran.chan")The compilation can take some time, so just be patient.
You'll also want to use a compiler that supports OpenMP to take advantage of parallelization.
Otherwise, the package can still be installed but will not be responsive to any setting of num.threads.
Once installed, usage is simple:
# Grabbing a test dataset.
library(scRNAseq)
x <- ZeiselBrainData()
mito <- grep("^mt-", rownames(x))
# Running the analysis to get the results.
out <- quickBasicAnalysis(assay(x), qc.subsets=list(Mt=mito))
# Optionally packaging into a SingleCellExperiment.
sce <- marshalToSCE(x, out, assay.type=1)Advanced users can use the various *.chan() functions to call specific steps.
Note that this requires a matrix to be initialized via initializeSparseMatrix(); they will not work with normal R matrices.
The scran.js package provides Javascript bindings to the same C++ libraries.
The scran-cli tool provides a command-line interface to a basic scRNA-seq analysis.
Add the quickMergedAnalysis() pipeline function.