1 Workflow version information

R version: R version 3.5.0 (2018-04-23)

Bioconductor version: 3.7

Package: 1.2.1

2 Introduction

Single-cell RNA sequencing (scRNA-seq) is widely used to measure the genome-wide expression profile of individual cells. From each cell, mRNA is isolated and reverse transcribed to cDNA for high-throughput sequencing (Stegle, Teichmann, and Marioni 2015). This can be done using microfluidics platforms like the Fluidigm C1 (Pollen et al. 2014), protocols based on microtiter plates like Smart-seq2 (Picelli et al. 2014), or droplet-based technologies like inDrop (Klein et al. 2015; Macosko et al. 2015). The number of reads mapped to each gene is then used to quantify its expression in each cell. Alternatively, unique molecular identifiers (UMIs) can be used to directly measure the number of transcript molecules for each gene (Islam et al. 2014). Count data are analyzed to detect highly variable genes (HVGs) that drive heterogeneity across cells in a population, to find correlations between genes and cellular phenotypes, or to identify new subpopulations via dimensionality reduction and clustering. This provides biological insights at a single-cell resolution that cannot be achieved with conventional bulk RNA sequencing of cell populations.

Strategies for scRNA-seq data analysis differ markedly from those for bulk RNA-seq. One technical reason is that scRNA-seq data are much noisier than bulk data (Brennecke et al. 2013; Marinov et al. 2014). Reliable capture (i.e., conversion) of transcripts into cDNA for sequencing is difficult with the low quantity of RNA in a single cell. This increases the frequency of drop-out events where none of the transcripts for a gene are captured. Dedicated steps are required to deal with this noise during analysis, especially during quality control. In addition, scRNA-seq data can be used to study cell-to-cell heterogeneity, e.g., to identify new cell subtypes, to characterize differentiation processes, to assign cells into their cell cycle phases, or to identify HVGs driving variability across the population (Vallejos, Marioni, and Richardson 2015; Fan et al. 2016; Trapnell et al. 2014). This is simply not possible with bulk data, meaning that custom methods are required to perform these analyses.

This article describes a set of computational workflows for basic analysis of scRNA-seq data, using software packages from the open-source Bioconductor project (Huber et al. 2015). The workflows start from a count matrix and describe the steps required for quality control to remove problematic cells; normalization of cell-specific biases, with and without spike-ins; cell cycle phase classification from gene expression data; data exploration to identify putative subpopulations; and finally, HVG and marker gene identification to prioritize interesting genes. The application of these procedures will be demonstrated on several public scRNA-seq datasets involving immortalized myeloid progenitors, brain cells, haematopoietic stem cells, T-helper cells and mouse embryonic stem cells, generated with a range of experimental protocols and platforms (Lun et al. 2017; Wilson et al. 2015; Zeisel et al. 2015; Islam et al. 2011; Buettner et al. 2015). The aim is to provide a variety of modular usage examples that can be applied to construct custom analysis pipelines.

Note: to cite this article, please refer to http://f1000research.com/articles/5-2122/v2 for instructions.

3 Links to individual workflows

This workflow will introduce most of the concepts for scRNA-seq data analysis, using a small dataset of 192 cells from an immortalized myeloid progenitor cell line (Lun et al. 2017) generated using the Smart-seq2 protocol (Picelli et al. 2014). It will start from loading of the data and finish at the characterization of cell clusters.

This workflow will adapt many of the previous concepts to analysis of UMI count data, using a large dataset of 3005 cells from the mouse brain (Zeisel et al. 2015) generated from the Fluidigm system (Pollen et al. 2014). It will focus on the modifications required to handle lower UMI counts and larger numbers of cells.

This workflow will adapt many of the previous concepts to analysis of droplet-based scRNA-seq data, using the publicly available dataset of 4000 peripheral blood mononuclear cells generated by 10X Genomics (Zheng et al. 2017). It will focus on the modifications required to handle cell calling, even lower counts and no spike-ins.

This workflow will describe some analyses that were not covered in the previous workflows, using a variety of relevant datasets (Wilson et al. 2015; Islam et al. 2011; Buettner et al. 2015) generated using a range of technologies. It includes alternative parameter settings and options that are not usually used as default but may be helpful in some circumstances.

This workflow will demonstrate an application of the mutual nearest neighbours method for batch correction, using several datasets from human pancreas (Grun et al. 2016; Muraro et al. 2016; Segerstolpe et al. 2016) generated with a range of technologies. It is based on analyses performed in Haghverdi et al. (2018), and will focus on the generation of corrected expression values that can be used in downstream analyses.

4 Obtaining a count matrix

All of these workflows start from a publicly available count matrix. For simplicity, we forego a description of the read processing steps required to generate the count matrix, i.e., read alignment and counting into features. These steps have been described in some detail elsewhere (Love et al. 2015; Chen, Lun, and Smyth 2016), and are largely the same for bulk and single-cell data. For users favouring an R-based approach to read alignment and counting, we suggest using the methods in the Rsubread package (Liao, Smyth, and Shi 2013, 2014).

Some considerations specific to single-cell data may also be relevant, depending on the experimental protocol used to generate the data:

• If spike-in RNA was added, the sequences of the spike-in transcripts can be included as additional FASTA files during genome index building prior to alignment. Similarly, genomic intervals for both spike-in transcripts and endogenous genes can be concatenated into a single GTF file prior to counting.
• If UMIs are present, these should be extracted from the read sequence prior to alignment, e.g., with the UMI-tools software (Smith, Heger, and Sudbery 2017). Reads with the same UMI mapping to the same gene represent a single underlying transcript molecule and only increment the count of that gene by one.
• Alignment-free methods such as kallisto (Bray et al. 2016) or Salmon (Patro, Duggal, and Kingsford 2015) may also be useful, due to their speed and ability to perform transcript quantification. This can be run using the functions runKallisto and runSalmon in the scater package.

5 Author information