Introduction

This vignette describes how to use the iSEEfier package to configure various initial states of iSEE instances, in order to simplify the task of visualizing single-cell RNA-seq, bulk RNA-seq data, or even your proteomics data in iSEE. In the remainder of this vignette, we will illustrate the main features of r BiocStyle::Biocpkg("iSEEfier") on a publicly available dataset from Baron et al. “A Single-Cell Transcriptomic Map of the Human and Mouse Pancreas Reveals Inter- and Intra-cell Population Structure”, published in Cell Systems in 2016. doi:10.1016/j.cels.2016.08.011. The data is made available via the scRNAseq Bioconductor package. We’ll simply use the mouse dataset, consisting of islets isolated from five C57BL/6 and ICR mice. # Getting started {#gettingstarted} To install iSEEfier package, we start R and enter:

if (!requireNamespace("BiocManager", quietly = TRUE))
  install.packages("BiocManager")
BiocManager::install("iSEEfier")

Once installed, the package can be loaded and attached to the current workspace as follows:

library("iSEEfier")

Create an initial state for gene expression visualization using iSEEinit()

When we have all input elements ready, we can create an iSEE initial state by running:

iSEEinit(sce = sce_obj,
         features = feature_list,
         reddim.type = reduced_dim,
         clusters = cluster,
         groups = group,
         add_markdown_panel = FALSE)

To configure the initial state of our iSEE instance using iSEEinit(), we need five parameters:

1. sce : A SingleCellExperiment object. This object stores information of different quantifications (counts, log-expression…), dimensionality reduction coordinates (t-SNE, UMAP…), as well as some metadata related to the samples and features. We’ll start by loading the sce object:

library("scRNAseq")
sce <- BaronPancreasData('mouse')
sce
#> class: SingleCellExperiment 
#> dim: 14878 1886 
#> metadata(0):
#> assays(1): counts
#> rownames(14878): X0610007P14Rik X0610009B22Rik ... Zzz3 l7Rn6
#> rowData names(0):
#> colnames(1886): mouse1_lib1.final_cell_0001 mouse1_lib1.final_cell_0002
#>   ... mouse2_lib3.final_cell_0394 mouse2_lib3.final_cell_0395
#> colData names(2): strain label
#> reducedDimNames(0):
#> mainExpName: NULL
#> altExpNames(0):

Let’s add the normalized counts

library("scuttle")
sce <- logNormCounts(sce)

Now we can add different dimensionality reduction coordinates

library("scater")
sce <- runPCA(sce)
sce <- runTSNE(sce)
sce <- runUMAP(sce)

Now our sce is ready, we can move on to the next argument.

2. features : which is a vector or a dataframe containing the genes/features of interest. Let’s say we would like to visualize the expression of some genes that were identified as marker genes for different cell population.

gene_list <- c("Gcg", # alpha
               "Ins1") # beta

3. reddim_type : In this example we decided to plot our data as a t-SNE plot.

reddim_type <- "TSNE"

4. clusters : Now we specify what clusters/cell-types/states/samples we would like to color/split our data with

# cell populations
cluster <- "label" #the name should match what's in the colData names

5. groups : Here we can add the groups/conditions/cell-types

# ICR vs C57BL/6
group <- "strain" #the name should match what's in the colData names

We can choose to include in this initial step a MarkdownBoard by setting the arguments add_markdown_panel to TRUE. At this point, all the elements are ready to be transferred into iSEEinit()

initial1 <- iSEEinit(sce = sce,
                    features = gene_list,
                    clusters = cluster,
                    groups = group,
                    add_markdown_panel = TRUE)

In case our features parameter was a data.frame, we could assign the name of the column containing the features to the gene_id parameter.

Now we are one step away from visualizing our list of genes of interest. All that’s left to do is to run iSEE with the initial state created with iSEEinit()

library("iSEE")
iSEE(sce, initial= initial1)

This instance, generated with iSEEinit(), returns a combination of panels, linked to each other, with the goal of visualizing the expression of certain marker genes in each cell population/group:

• A ReducedDimensionPlot, FeatureAssayPlot and RowDataTable for each single gene in features.
• A ComplexHeatmapPlot with all genes in features
• A ColumnDataPlot panel
• A MarkdownBoard panel

Create an initial state for feature sets exploration using iSEEnrich()

Sometimes it is interesting to look at some specific feature sets and the associated genes. That’s when the utility of iSEEnrich becomes apparent. We will need 4 elements to explore feature sets of interest:

• sce: A SingleCellExperiment object
• collection: A character vector specifying the gene set collections of interest (it is possible to use GO or KEGG terms)
• gene_identifier: A character string specifying the identifier to use to extract gene IDs for the organism package. This can be “ENS” for ENSEMBL ids, “SYMBOL” for gene names…
• organism: A character string of the org.*.eg.db package to use to extract mappings of gene sets to gene IDs.
• reddim_type: A string vector containing the dimensionality reduction type
• clusters: A character string containing the name of the clusters/cell-type/state…(as listed in the colData of the sce)
• groups: A character string of the groups/conditions…(as it appears in the colData of the sce)

GO_collection <- "GO"
Mm_organism <- "org.Mm.eg.db"
gene_id <- "SYMBOL"
cluster <- "label"
group <- "strain"
reddim_type <- "PCA"

Now let’s create this initial setup for iSEE using iSEEnrich()

results <- iSEEnrich(sce = sce,
                     collection = GO_collection,
                     organism = Mm_organism,
                     gene_identifier = gene_id,
                     clusters = cluster,
                     groups = group,
                     reddim_type = reddim_type)

iSEEnrich will specifically return a list with the updated sce object and its associated initial configuration. To start the iSEE instance we run:

iSEE(results$sce, initial = results$initial)

Create an initial state for marker gene exploration using iSEEmarker()

In many cases, we are interested in determining the identity of our clusters, or further subset our cells types. That’s where iSEEmarker() comes in handy. Similar to iSEEinit(), we need the following parameters:

• sce: a SingleCellExperiment object
• clusters: the name of the clusters/cell-type/state
• groups: the groups/conditions
• selection_plot_format: the class of the panel that we will be using to select the clusters of interest.

initial3 <- iSEEmarker(
  sce = sce,
  clusters = cluster,
  groups = group,
  selection_plot_format = "ColumnDataPlot")

This function returns a list of panels, with the goal of visualizing the expression of marker genes selected from the DynamicMarkerTable in each cell cell type. Unlike iSEEinit(), which requires us to specify a list of genes, iSEEmarker() utilizes the DynamicMarkerTable that performs statistical testing through the findMarkers() function from the scran package. To start exploring the marker genes of each cell type with iSEE, we run:

iSEE(sce, initial = initial3)

Visualize a preview of the initial configurations with view_initial_tiles()

Previously, we successfully generated three distinct initial configurations for iSEE. However, understanding the expected content of our iSEE instances is not always straightforward. That’s when we can use view_initial_tiles(). We only need as an input the initial configuration to obtain a graphical visualization of the expected the corresponding iSEE instance:

library(ggplot2)
view_initial_tiles(initial = initial1)

view_initial_tiles(initial = results$initial)

Visualize network connections between panels with view_initial_network()

As some of these panels are linked to each other, we can visualize these networks with view_initial_network(). Similar to iSEEconfigviewer(), this function takes the initial setup as input: This function always returns the igraph object underlying the visualizations that can be displayed as a side effect.

library("igraph")
library("visNetwork")
g1 <- view_initial_network(initial1, plot_format = "igraph")

g1
#> IGRAPH bf881c9 DN-- 11 3 -- 
#> + attr: name (v/c), color (v/c)
#> + edges from bf881c9 (vertex names):
#> [1] ReducedDimensionPlot1->ColumnDataPlot1  
#> [2] ReducedDimensionPlot2->ColumnDataPlot1  
#> [3] ReducedDimensionPlot3->FeatureAssayPlot3
initial2 <- results$initial
g2 <- view_initial_network(initial2, plot_format = "visNetwork")

Merge different initial configurations with glue_initials()

Sometimes, it would be interesting to merge different iSEE initial configurations to visualize all different panel in the same iSEE instance.

merged_config <- glue_initials(initial1,initial2)

We can then preview the content of this initial configuration

view_initial_tiles(merged_config)

Related work

The idea of launching iSEE() with some specific configuration is not entirely new, and it was covered in some use cases by the mode_ functions available in the iSEEu package. There, the user has access to the following:

• iSEEu::modeEmpty() - this will launch iSEE without any panels, and let you build up the configuration from the scratch. Easy to start, easy to build.
• iSEEu::modeGating() - this will open iSEE with multiple chain-linked FeatureExpressionPlot panels, just like when doing some in silico gating. This could be a very good fit if working with mass cytometry data.
• iSEEu::modeReducedDim() - iSEE will be ready to compare multiple ReducedDimensionPlot panels, which is a suitable option to compare the views resulting from different embeddings (and/or embeddings generated with slightly different parameter configurations). The modes directly launch an instance of iSEE, whereas the functionality in iSEEfier is rather oriented to obtain more tailored-to-the-data-at-hand initial objects, that can subsequently be passed as an argument to the iSEE() call. We encourage users to submit suggestions about their “classical ways” of using iSEE on their data - be that by opening an issue or already proposing a Pull Request on GitHub.