scater: single-cell analysis toolkit for expression with R

[//]: # () [//]: # ()

This package contains useful tools for the analysis of single-cell gene expression data using the statistical software R. The package places an emphasis on tools for quality control, visualisation and pre-processing of data before further downstream analysis.

We hope that scater fills a useful niche between raw RNA-sequencing count or transcripts-per-million data and more focused downstream modelling tools such as monocle, scLVM, SCDE, edgeR, limma and so on.

Briefly, scater enables the following:

1. Automated computation of QC metrics
2. Transcript quantification from read data with pseudo-alignment
3. Data format standardisation
4. Rich visualisations for exploratory analysis
5. Seamless integration into the Bioconductor universe
6. Simple normalisation methods

See below for information about installation, getting started and highlights of the package.

Installation

Installation from Bioconductor (recommended)

The scater package has been accepted into Bioconductor! Thus, the most reliable way to install the package is to use the usual Bioconductor method:

## try http:// if https:// URLs are not supported
source("https://bioconductor.org/biocLite.R")
biocLite("scater")

Currently, only the "devel" (i.e. development) version of scater is available through Bioconductor. This means that you will need to be using Bioconductor devel and the development version of R (R 3.3) in order to install scater from Bioconductor.

The scater package will become available as a "release" version in the next Bioconductor release in April 2016. At this point the release version of scater will work with the release version of R and Bioconductor, and development will continue in the devel version of the package.

Installation from Github

Alternatively, scater can be installed directly from GitHub as described below. In this case, package that scater uses ("depends on" in R parlance) will not be automatically installed, so you will have to install the required packages as shown below.

I recommend using Hadley Wickham's devtools package to install scater directly from GitHub. If you don't have devtools installed, then install that from CRAN (as shown below) and then run the call to install scater:

If you are using the development version of R, 3.3:

install.packages("devtools")
devtools::install_github("davismcc/scater", build_vignettes = TRUE)

If you are using the current release version of R, 3.2.3:

devtools::install_github("davismcc/scater", ref = "release-R-3.2", build_vignettes = TRUE)

If you find that the above will not install on Linux systems, please try with the option build_vignettes = FALSE. This is a known issue that we are working to resolve.

Using the most recent version of R is strongly recommended (R 3.2.3 at the time of writing). Effort has been made to ensure the package works with R >3.0, but the package has not been tested with R <3.1.1.

There are several other packages from CRAN and Bioconductor that scater uses, so you will need to have these packages installed as well. The CRAN packages should install automatically when scater is installed, but you will need to install the Bioconductor packages manually.

Not all of the following are strictly necessary, but they enhance the functionality of scater and are good packages in their own right. The commands below should help with package installations.

CRAN packages:

install.packages(c("data.table", "ggplot2", "knitr", "matrixStats", "MASS",
                "plyr", "reshape2", "rjson", "testthat", "viridis"))

Bioconductor packages:

source("http://bioconductor.org/biocLite.R")
biocLite(c("Biobase", "biomaRt", "edgeR", "limma", "rhdf5"))

Optional packages that are not strictly required but enhance the functionality of scater:

install.packages(c("cowplot", "cluster", "mvoutlier", "parallel", "Rtsne"))
biocLite(c("destiny", "monocle"))

You might also like to install dplyr for convenient data manipulation:

install.packages("dplyr")

Getting started

The best place to start is the vignette. From inside an R session, load scater and then browse the vignettes:

library(scater)
browseVignettes("scater")

There is a detailed HTML document available that introduces the main features and functionality of scater.

scater workflow

The diagram below provised an overview of the pre-processing and QC workflow possible in scater, listing the functions that can be used at various stages.

Highlights

The scater package allows you to do some neat things relatively quickly. Some highlights are shown below with example code and screenshots.

1. Automated computation of QC metrics
2. Transcript quantification from read data with pseudo-alignment approaches
3. Data format standardisation
4. Rich visualisations for QC and exploratory analysis
5. Seamless integration into the Bioconductor universe
6. Simple normalisation methods

For details of how to use these functions, please consult the vignette and package documentation. The plots shown use the example data included with the package (for which there is no interesting structure) and as shown require only one or two lines of code to generate.

Automatic computation of QC metrics

Use the calculateQCMetrics function to compute many metrics useful for gene/transcript-level and cell-level QC. Metrics computed include number of genes expressed per cell, percentage of expression from control genes (e.g. ERCC spike-ins) and many more.

Transcript quantification with kallisto

The runKallisto function provides a wrapper to the kallisto software for quantifying transcript abundance from FASTQ files using a pseudo-alignment approach. This new approach is extremely fast. With readKallisto, transcript quantities can be read into a data object in R.

Plotting functions

Default plot for an SCESet object gives cumulative expression for the most-expressed features (genes or transcripts)

The plotTSNE function produces a t-distributed stochastic neighbour embedding plot for the cells.

The plotPCA function produces a principal components analysis plot for the cells.

The plotDiffusionMap function produces a diffusion map plot for the cells.

The plotExpression function plots the expression values for a selection of features.

The plotQC function produces a variety of QC plots useful for diagnostics and feature and cell filtering. It can be used to plot the most highly-expressed genes (or features) in the data set or create density plots to assess the relative importance of explanatory variables, as well as many other visualisations useful for QC.

The plotPhenoData function plots two phenotype metadata variables (such as QC metrics).

See also plotFeatureData to plot feature (gene) metadata variables, including QC metrics.

Plus many, many more possibilities. Please consult the vignette and documentation for details.

Acknowledgements and disclaimer

The package leans heavily on previously published work and packages, namely edgeR and limma. The SCESet class is inspired by the CellDataSet class from monocle, and SCESet objects in scater can be easily converted to and from monocle's CellDataSet objects.

The package is currently in an Beta state. The major functionality of the package is settled, but it is still under development so may change from time to time. Please do try it and contact me with bug reports, feedback, feature requests, questions and suggestions to improve the package.

Davis McCarthy, February 2016