1 Introduction

In the single cell World, which includes flow cytometry, mass cytometry, single-cell RNA-seq (scRNA-seq), and others, there is a need to improve data visualisation and to bring analysis capabilities to researchers even from non-technical backgrounds. scDataviz (Blighe 2020) attempts to fit into this space, while also catering for advanced users. Additonally, due to the way that scDataviz is designed, which is based on SingleCellExperiment (Lun and Risso 2020), it has a ‘plug and play’ feel, and immediately lends itself as flexibile and compatibile with studies that go beyond scDataviz. Finally, the graphics in scDataviz are generated via the ggplot (Wickham 2016) engine, which means that users can ‘add on’ features to these with ease.

This package just provides some additional functions for dataviz and clustering, and provides another way of identifying cell-types in clusters. It is not strictly intended as a standalone analysis package. For a comprehensive high-dimensional cytometry workflow, it is recommended to check out the work by Nowicka et al. CyTOF workflow: differential discovery in high-throughput high-dimensional cytometry datasets. For a more comprehensive scRNA-seq workflow, please check out OSCA and Analysis of single cell RNA-seq data.

2 Installation

2.2 2. Load the package into R session

3 Tutorial 1: CyTOF FCS data

Here, we will utilise some of the flow cytometry data from Deep phenotyping detects a pathological CD4+ T-cell complosome signature in systemic sclerosis.

This can normally be downloadedd via git clone from your command prompt:

In a practical situation, we would normally read in this data from the raw FCS files and then QC filter, normalise, and transform them. This can be achieved via the processFCS function, which, by default, also removes variables based on low variance and downsamples [randomly] your data to 100000 variables. The user can change these via the downsample and downsampleVar parameters. An example (not run) is given below:

In flow and mass cytometry, getting the correct marker names in the FCS files can be surprisingly difficult. In many cases, from experience, a facility may label the markers by their metals, such as Iridium (Ir), Ruthenium (Ru), Terbium (Tb), et cetera - this is the case for the data used in this tutorial. The true marker names may be held as pData encoded within each FCS, accessible via:

Whatever the case, it is important to sort out marker naming issues prior to the experiment being conducted in order to avoid any confusion.

For this vignette, due to the fact that the raw FCS data is > 500 megabytes, we will work with a smaller pre-prepared dataset that has been downsampled to 10000 cells using the above code. This data comes included with the package.

Load the pre-prepared complosome data.

One can also create a new SingleCellExperiment object manually using any type of data, including any data from scRNA-seq produced elsewhere. Import functions for data deriving from other sources is covered in Tutorials 2 and 3 in this vignette. All functions in scDataviz additionally accept data-frames or matrices on their own, de-necessitating the reliance on the SingleCellExperiment class.

3.2 Perform UMAP

UMAP can be performed on the entire dataset, if your computer’s memory will permit. Currently it’s default is to use the data contained in the ‘scaled’ assay component of your SingleCellExperiment object.

UMAP can also be stratified based on a column in your metadata, e.g., (treated versus untreated samples); however, to do this, I recommend creating separate SingleCellExperiment objects from the very start, i.e., from the the data input stage, and processing the data separately for each group.

Nota bene - advanced users may want to change the default configuration for UMAP. scDataviz currently performs UMAP via the umap package. In order to modify the default configuration, one can pull in the default config separately from the umap package and then modify these config values held in the umap.defaults variable, as per the umap vignette (see ‘Tuning UMAP’ section). For example:

We can also perform UMAP on a select number of PC eigenvectors. PCAtools (Blighe and Lun 2020) can be used to infer ideal number of dimensions to use via the elbow method and Horn’s parallel analysis.

## PC3 
##   3
## [1] 1

For now, let’s just use 5 PCs.

3.3 Create a contour plot of the UMAP layout

This and the remaining sections in this tutorial are about producing great visualisations of the data and attempting to make sense of it, while not fully overlapping with functionalioty provided by other programs that operate in tis space.

With the contour plot, we are essentially looking at celluar density. It can provide for a beautiful viusualisation in a manuscript while also serving as a useful QC tool: if the density is ‘scrunched up’ into a single area in the plot space, then there are likely issues with your input data distribution. We want to see well-separated, high density ‘islands’, or, at least, gradual gradients that blend into one another across high density ‘peaks’.

Create a contour plot of the UMAP layout

Create a contour plot of the UMAP layout

3.5 Shade cells by metadata

Shading cells by metadata can be useful for identifying any batch effects, but also useful for visualising, e.g., differences across treatments.

First, let’s take a look inside the metadata that we have.

##       sample   group treatment
## cell1    P00 Disease    Unstim
## cell2    P00 Disease    Unstim
## cell3    P04 Disease      CD46
## cell4    P03 Disease      CD46
## cell5    P08 Disease    Unstim
## cell6    P00 Disease      CD46
## [1] "Healthy" "Disease"
## [1] "CD46"   "Unstim" "CD3"
Shade cells by metadata

Shade cells by metadata