In stemangiola/tidySCE: Brings SingleCellExperiment to the Tidyverse

library(knitr)
knitr::opts_chunk$set(
    cache=TRUE, warning=FALSE,
    message=FALSE, cache.lazy=FALSE)

Introduction {-}

tidySingleCellExperiment provides a bridge between Bioconductor single-cell packages [@amezquita2019orchestrating] and the tidyverse [@wickham2019welcome]. It enables viewing the Bioconductor r BiocStyle::Biocpkg("SingleCellExperiment") object as a tidyverse tibble, and provides SingleCellExperiment-compatible r BiocStyle::CRANpkg("dplyr"), r BiocStyle::CRANpkg("tidyr"), r BiocStyle::CRANpkg("ggplot2") and r BiocStyle::CRANpkg("plotly") functions (see Table \@ref(tab:table)). This allows users to get the best of both Bioconductor and tidyverse worlds.

| ------ | ---------- All functions compatible with SingleCellExperiments | After all, a tidySingleCellExperiment
is a SingleCellExperiment, just better! tidyverse |
dplyr | All tibble-compatible
functions (e.g., select()) tidyr | All tibble-compatible
functions (e.g., pivot_longer()) ggplot2 | Plotting with ggplot() plotly | Plotting with plot_ly() Utilities |
as_tibble() | Convert cell-wise information to a tbl_df join_features() | Add feature-wise information;
returns a tbl_df aggregate_cells() | Aggregate feature abundances as pseudobulks;
returns a SummarizedExperiment : (#tab:table) Available tidySingleCellExperiment functions and utilities.

Installation {-}

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

BiocManager::install("tidySingleCellExperiment")

Load libraries used in this vignette.

# Bioconductor single-cell packages
library(scran)
library(scater)
library(igraph)
library(celldex)
library(SingleR)
library(SingleCellSignalR)

# Tidyverse-compatible packages
library(purrr)
library(GGally)
library(tidyHeatmap)

# Both
library(tidySingleCellExperiment)

# Other
library(Matrix)
library(dittoSeq)

Data representation of tidySingleCellExperiment

This is a SingleCellExperiment object but it is evaluated as a tibble. So it is compatible both with SingleCellExperiment and tidyverse.

data(pbmc_small, package="tidySingleCellExperiment")
pbmc_small_tidy <- pbmc_small

It looks like a tibble...

pbmc_small_tidy

...but it is a SingleCellExperiment after all!

counts(pbmc_small_tidy)[1:5, 1:4]

The SingleCellExperiment object's tibble visualisation can be turned off, or back on at any time.

# Turn off the tibble visualisation
options("restore_SingleCellExperiment_show" = TRUE)
pbmc_small_tidy

# Turn on the tibble visualisation
options("restore_SingleCellExperiment_show" = FALSE)

Annotation polishing

We may have a column that contains the directory each run was taken from, such as the "file" column in pbmc_small_tidy.

pbmc_small_tidy$file[1:5]

We may want to extract the run/sample name out of it into a separate column. The tidyverse function extract() can be used to convert a character column into multiple columns using regular expression groups.

# Create sample column
pbmc_small_polished <-
    pbmc_small_tidy %>%
    extract(file, "sample", "../data/([a-z0-9]+)/outs.+", remove=FALSE)

# Reorder to have sample column up front
pbmc_small_polished %>%
    select(sample, everything())

Preliminary plots

Set colours and theme for plots.

# Use colourblind-friendly colours
friendly_cols <- dittoSeq::dittoColors()

# Set theme
custom_theme <- list(
    scale_fill_manual(values=friendly_cols),
    scale_color_manual(values=friendly_cols),
    theme_bw() + theme(
        aspect.ratio=1,
        legend.position="bottom",
        axis.line=element_line(),
        text=element_text(size=12),
        panel.border=element_blank(),
        strip.background=element_blank(),
        panel.grid.major=element_line(linewidth=0.2),
        panel.grid.minor=element_line(linewidth=0.1),
        axis.title.x=element_text(margin=margin(t=10, r=10, b=10, l=10)),
        axis.title.y=element_text(margin=margin(t=10, r=10, b=10, l=10))))

We can treat pbmc_small_polished as a tibble for plotting.

Here we plot number of features per cell.

pbmc_small_polished %>%
    ggplot(aes(nFeature_RNA, fill=groups)) +
    geom_histogram() +
    custom_theme

Here we plot total features per cell.

pbmc_small_polished %>%
    ggplot(aes(groups, nCount_RNA, fill=groups)) +
    geom_boxplot(outlier.shape=NA) +
    geom_jitter(width=0.1) +
    custom_theme

Here we plot abundance of two features for each group.

pbmc_small_polished %>%
    join_features(features=c("HLA-DRA", "LYZ")) %>%
    ggplot(aes(groups, .abundance_counts + 1, fill=groups)) +
    geom_boxplot(outlier.shape=NA) +
    geom_jitter(aes(size=nCount_RNA), alpha=0.5, width=0.2) +
    scale_y_log10() +
    custom_theme

Preprocessing

We can also treat pbmc_small_polished as a SingleCellExperiment object and proceed with data processing with Bioconductor packages, such as r BiocStyle::Biocpkg("scran") [@lun2016pooling] and r BiocStyle::Biocpkg("scater") [@mccarthy2017scater].

# Identify variable genes with scran
variable_genes <-
    pbmc_small_polished %>%
    modelGeneVar() %>%
    getTopHVGs(prop=0.1)

# Perform PCA with scater
pbmc_small_pca <-
    pbmc_small_polished %>%
    runPCA(subset_row=variable_genes)

pbmc_small_pca

If a tidyverse-compatible package is not included in the tidySingleCellExperiment collection, we can use as_tibble() to permanently convert a tidySingleCellExperiment into a tibble.

# Create pairs plot with 'GGally'
pbmc_small_pca %>%
    as_tibble() %>%
    select(contains("PC"), everything()) %>%
    GGally::ggpairs(columns=1:5, aes(colour=groups)) +
    custom_theme

Clustering

We can proceed with cluster identification with r BiocStyle::Biocpkg("scran").

pbmc_small_cluster <- pbmc_small_pca

# Assign clusters to the 'colLabels'
# of the 'SingleCellExperiment' object
colLabels(pbmc_small_cluster) <-
    pbmc_small_pca %>%
    buildSNNGraph(use.dimred="PCA") %>%
    igraph::cluster_walktrap() %$%
    membership %>%
    as.factor()

# Reorder columns
pbmc_small_cluster %>%
    select(label, everything())

And interrogate the output as if it was a regular tibble.

# Count number of cells for each cluster per group
pbmc_small_cluster %>%
    count(groups, label)

We can identify and visualise cluster markers combining SingleCellExperiment, tidyverse functions and r BiocStyle::CRANpkg("tidyHeatmap") [@mangiola2020tidyheatmap].

# Identify top 10 markers per cluster
marker_genes <-
    pbmc_small_cluster %>%
    findMarkers(groups=pbmc_small_cluster$label) %>%
    as.list() %>%
    map(~ .x %>%
        head(10) %>%
        rownames()) %>%
    unlist()

# Plot heatmap
pbmc_small_cluster %>%
    join_features(features=marker_genes) %>%
    group_by(label) %>%
    heatmap(
        .row=.feature, .column=.cell, 
        .value=.abundance_counts, scale="column")

Reduce dimensions

We can calculate the first 3 UMAP dimensions using r BiocStyle::Biocpkg("scater").

pbmc_small_UMAP <-
    pbmc_small_cluster %>%
    runUMAP(ncomponents=3)

And we can plot the result in 3D using r BiocStyle::CRANpkg("plotly").

pbmc_small_UMAP %>%
    plot_ly(
        x=~`UMAP1`,
        y=~`UMAP2`,
        z=~`UMAP3`,
        color=~label,
        colors=friendly_cols[1:4])

Cell type prediction

We can infer cell type identities using r BiocStyle::Biocpkg("SingleR") [@aran2019reference] and manipulate the output using tidyverse.

# Get cell type reference data
blueprint <- celldex::BlueprintEncodeData()

# Infer cell identities
cell_type_df <- 
    logcounts(pbmc_small_UMAP) %>%
    Matrix::Matrix(sparse = TRUE) %>%
    SingleR::SingleR(
        ref=blueprint,
        labels=blueprint$label.main,
        method="single") %>%
    as.data.frame() %>%
    as_tibble(rownames="cell") %>%
    select(cell, first.labels)

# Join UMAP and cell type info
data(cell_type_df)
pbmc_small_cell_type <-
    pbmc_small_UMAP %>%
    left_join(cell_type_df, by="cell")

# Reorder columns
pbmc_small_cell_type %>%
    select(cell, first.labels, everything())

We can easily summarise the results. For example, we can see how cell type classification overlaps with cluster classification.

# Count number of cells for each cell type per cluster
pbmc_small_cell_type %>%
    count(label, first.labels)

We can easily reshape the data for building information-rich faceted plots.

pbmc_small_cell_type %>%
    # Reshape and add classifier column
    pivot_longer(
        cols=c(label, first.labels),
        names_to="classifier", values_to="label") %>%
    # UMAP plots for cell type and cluster
    ggplot(aes(UMAP1, UMAP2, color=label)) +
    facet_wrap(~classifier) +
    geom_point() +
    custom_theme

We can easily plot gene correlation per cell category, adding multi-layer annotations.

pbmc_small_cell_type %>%
    # Add some mitochondrial abundance values
    mutate(mitochondrial=rnorm(dplyr::n())) %>%
    # Plot correlation
    join_features(features=c("CST3", "LYZ"), shape="wide") %>%
    ggplot(aes(CST3+1, LYZ+1, color=groups, size=mitochondrial)) +
    facet_wrap(~first.labels, scales="free") +
    geom_point() +
    scale_x_log10() +
    scale_y_log10() +
    custom_theme

Nested analyses

A powerful tool we can use with tidySingleCellExperiment is tidyverse's nest(). We can easily perform independent analyses on subsets of the dataset. First, we classify cell types into lymphoid and myeloid, and then nest() based on the new classification.

pbmc_small_nested <-
    pbmc_small_cell_type %>%
    filter(first.labels != "Erythrocytes") %>%
    mutate(cell_class=if_else(
        first.labels %in% c("Macrophages", "Monocytes"),
        true="myeloid", false="lymphoid")) %>%
    nest(data=-cell_class)

pbmc_small_nested

Now we can independently for the lymphoid and myeloid subsets (i) find variable features, (ii) reduce dimensions, and (iii) cluster using both tidyverse and SingleCellExperiment seamlessly.

pbmc_small_nested_reanalysed <-
    pbmc_small_nested %>%
    mutate(data=map(data, ~ {
        # feature selection
        variable_genes <- .x %>%
            modelGeneVar() %>%
            getTopHVGs(prop=0.3)
        # dimension reduction
        .x <- .x %>%
            runPCA(subset_row=variable_genes) %>%
            runUMAP(ncomponents=3)
        # clustering
        colLabels(.x) <- .x %>%
            buildSNNGraph(use.dimred="PCA") %>%
            cluster_walktrap() %$%
            membership %>%
            as.factor()
        return(.x)
    }))
pbmc_small_nested_reanalysed

We can then unnest() and plot the new classification.

pbmc_small_nested_reanalysed %>%
    # Convert to 'tibble', else 'SingleCellExperiment'
    # drops reduced dimensions when unifying data sets.
    mutate(data=map(data, ~as_tibble(.x))) %>%
    unnest(data) %>%
    # Define unique clusters
    unite("cluster", c(cell_class, label), remove=FALSE) %>%
    # Plotting
    ggplot(aes(UMAP1, UMAP2, color=cluster)) +
    facet_wrap(~cell_class) +
    geom_point() +
    custom_theme

We can perform a large number of functional analyses on data subsets. For example, we can identify intra-sample cell-cell interactions using SingleCellSignalR [@cabello2020singlecellsignalr], and then compare whether interactions are stronger or weaker across conditions. The code below demonstrates how this analysis could be performed. It won't work with this small example dataset as we have just two samples (one for each condition). But some example output is shown below and you can imagine how you can use tidyverse on the output to perform t-tests and visualisation.

pbmc_small_nested_interactions <-
    pbmc_small_nested_reanalysed %>%
    # Unnest based on cell category
    unnest(data) %>%
    # Create unambiguous clusters
    mutate(integrated_clusters=first.labels %>% as.factor() %>% as.integer()) %>%
    # Nest based on sample
    nest(data=-sample) %>%
    mutate(interactions=map(data, ~ {
        # Produce variables. Yuck!
        cluster <- colData(.x)$integrated_clusters
        data <- data.frame(assay(.x) %>% as.matrix())
        # Ligand/Receptor analysis using 'SingleCellSignalR'
        data %>%
            cell_signaling(genes=rownames(data), cluster=cluster) %>%
            inter_network(data=data, signal=., genes=rownames(data), cluster=cluster) %$%
            `individual-networks` %>%
            map_dfr(~ bind_rows(as_tibble(.x)))
    }))

pbmc_small_nested_interactions %>%
    select(-data) %>%
    unnest(interactions)

If the dataset was not so small, and interactions could be identified, you would see something like below.

data(pbmc_small_nested_interactions)
pbmc_small_nested_interactions

Aggregating cells

Sometimes, it is necessary to aggregate the gene-transcript abundance from a group of cells into a single value. For example, when comparing groups of cells across different samples with fixed-effect models.

In tidySingleCellExperiment, cell aggregation can be achieved using aggregate_cells(), which will return an object of class r BiocStyle::Biocpkg("SummarizedExperiment").

pbmc_small_tidy %>%
  aggregate_cells(groups, assays="counts")

Session Info

sessionInfo()

References

stemangiola/tidySCE documentation built on Sept. 22, 2024, 10:05 p.m.