Skip to content

Lecture notes. Written from the material of StatOmics Sga21, licensed CC BY-NC-SA 4.0. These are notes, not a transcript: the material has been reorganised and rewritten. This adaptation carries the same licence, and the original is linked above.

36. Single-Cell RNA-Seq: A First Look

What this covers

This chapter is the first hands-on encounter with single-cell RNA-sequencing (scRNA-seq) data in the course: loading a real scRNA-seq dataset in R/Bioconductor, inspecting how it is stored, and turning the same visualisation tools already used for bulk RNA-seq — PCA and multidimensional scaling (MDS) — on it to see how well they recover known biological structure. It assumes the reader already knows how to build and read a PCA or MDS plot from a bulk RNA-seq count matrix, and is comfortable with the standard Bioconductor container classes for expression data.

The dataset: a mouse retina, one cell at a time

The practical works with MacoskoRetinaData(), a dataset shipped in the Bioconductor package scRNAseq. Where a bulk RNA-seq experiment gives one expression profile per sample — a pool of many cells averaged together — this dataset gives one expression profile per individual mouse retina cell.

if(!"BiocManager" %in% installed.packages()[,1]){
  install.packages("BiocManager")
}
if(!"scRNAseq" %in% installed.packages()[,1]){
  BiocManager::install("scRNAseq")
}

suppressPackageStartupMessages(library(scRNAseq))
sce <- MacoskoRetinaData()
sce
class(sce)
counts(sce)[1:5,1:5]
head(colData(sce))

# filter cells
sce <- sce[,!is.na(colData(sce)$cluster)]
sce

sce is returned as a SingleCellExperiment, the standard Bioconductor container for single-cell data. It behaves the way a SummarizedExperiment does for bulk data, with one change that matters for everything that follows:

  • counts(sce) is the expression matrix, but its columns are now individual cells rather than bulk samples — printing counts(sce)[1:5,1:5] shows the first five genes against the first five cells.
  • colData(sce) is the per-column (per-cell) metadata. Here it carries a cluster column: a label of which cell type or cluster each cell belongs to.
  • Not every cell has a cluster label. The line sce <- sce[,!is.na(colData(sce)$cluster)] drops the cells whose cluster entry is NA, keeping only the labelled ones — which is what makes it possible, later, to colour a plot by cluster and check whether the plot's structure matches the label.

That cluster column is the ground truth the rest of the exercise checks itself against: if a visualisation method is doing its job, cells that share a label should end up close together, and cells with different labels should end up apart.

Exercises

  • Explore the sce object directly — its class, its dimensions, the count matrix, and the cell metadata. Compared with a bulk RNA-seq count matrix, what is structurally different about this dataset?
  • Build a PCA plot and an MDS plot of the cells, using the same tools used earlier in the course for bulk RNA-seq. Colour the points by the cluster label found in colData(sce).
  • Try visualising the structure of the dataset with any other tool you want.
  • Having tried the above: do your plots recapitulate the known structure — do the cell-type clusters separate out? Where do you run into trouble, and why do you think that happens?

Sources

  • docs/omics-statistics/statomics/sga21/singleCell_intro1.md — converted from singleCell_intro1.Rmd (statOmics SGA21 course, CC BY-NC-SA 4.0). This is the only material supplied for this chapter: no slides, transcript, or separate problem set were given, and the chapter's code and exercises reproduce the practical in full.

← 35. Filtering, Aliasing, and limma-voom · Contents · 38. Single-Cell QC and Dimensionality Reduction →