Evaluating the performance of different matrix types

Overview

This repository contains code for the paper beachmat: a Bioconductor C++ API for accessing high-throughput biological data from a variety of R matrix types by Lun et al. (2018).

The provided code will check the performance of different matrix types for row/column access, using simulated and real data sets. To run the tests on your machine, please read the following instructions.

Setup

Install beachmat from Bioconductor.
Enter timings and run R CMD INSTALL --clean package. This requires installation of RcppArmadillo and RcppEigen.

Running simulations

timings/ contains scripts for timings (in milliseconds) for accessing data from different matrix representations.
timings/chunking/ contains scripts for timing rechunking, as well as checking the chunk cache logic.
memory/ contains scripts for memory usage for different matrix representations.
miscellaneous contains scripts to compare timings to R, and to verify the no-copy access method of RcppArmadillo and RcppEigen.

Running real analyses

Zeisel dataset

Enter real/zeisel and download the count matrix for the Zeisel data set.

Execute the zeisel_time.R script to generate timings (in milliseconds) for matrix access to this data. This will also determine memory usage for each matrix representation.
Execute the detection_stats.R script to generate timings (in milliseconds) for computing various cell- or gene-based statistics from this data.

10X dataset

Enter real/10X and install TENxBrainData. Read the README.md file for order of evaluation of the various Rmarkdown scripts.