GraphExperiment S4 class

The GraphExperiment class was designed to represent rectangular, quantitative data (e.g., from transcriptomics, proteomics, metabolomics) along with graphs showing how features (e.g., genes, proteins, compounds) and observations (e.g., samples, cells, species, spots) interact with each other. It extends SingleCellExperiment by providing users with additional slots where row/column graphs can be stored.

Usage

GraphExperiment(..., rowGraphs = list(), colGraphs = list())

Arguments

...

Arguments passed to the SingleCellExperiment constructor function.

rowGraphs

A list of igraph objects with one or multiple graphs representing how features relate to each other. Node names (i.e., V(graph)$name) must match rownames of assays.

colGraphs

A list of igraph objects with one or multiple graphs representing how observations relate to each other. Node names (i.e., V(graph)$name) must match colnames of assays.

Value

A GraphExperiment object.

Details

Like SingleCellExperiment, the GraphExperiment S4 class stores quantitative data with associated metadata (i.e., rowData and colData) along with embeddings from dimensionality reduction techniques. However, it provides users with additional containers for igraph objects containing graphs describing how features and/or observations interact with each other. Graphs for features are stored in a rowGraphs slot, and graphs for observations are stored in a colGraphs slot. Both slots can hold one or multiple igraph objects with some sort of network representation of the features in rownames or observations in colnames. Example graphs for features can be coexpression networks, regulatory networks, or co-abundance networks. Example graphs for observations can be cell-cell distances, sample-sample distances, or species networks representing genealogies (as an alternative to phylogenies).

Importantly, node names in each row graph must match rownames of the assays, and node names in each column graph must match colnames of the assays. Besides, subsetting methods simultaneously subset assays, rowData, rowGraphs, colData, and colGraphs.

Besides the constructor function (GraphExperiment()), a GraphExperiment object can also be created by coercing from a SummarizedExperiment or SingleCellExperiment object.

Examples

# Example 1: from constructor function ----
## Simulate a matrix with 200 genes and 100 cells
gene_ids <- paste0("gene", seq_len(200))
cell_ids <- paste0("cell", seq_len(100))
mat <- matrix(rpois(20000, 5), ncol = 100, dimnames = list(gene_ids, cell_ids))

## Create a rowGraph from correlations (`igraph` object)
g <- graph_from_adjacency_matrix(cor(t(mat)), weighted = TRUE)

## Create a colGraph, also from correlations (but see scran::buildSNNGraph)
g2 <- graph_from_adjacency_matrix(cor(mat), weighted = TRUE)

## Construct `GraphExperiment` object
ge <- GraphExperiment(
    assays = list(counts = mat),
    rowGraphs = list(cor = g),
    colGraphs = list(cellcor = g2)
)
ge
#> class: GraphExperiment 
#> dim: 200 100 
#> metadata(0):
#> assays(1): counts
#> rownames(200): gene1 gene2 ... gene199 gene200
#> rowData names(0):
#> colnames(100): cell1 cell2 ... cell99 cell100
#> colData names(0):
#> reducedDimNames(0):
#> mainExpName: NULL
#> altExpNames(0):
#> rowGraphs(1): cor
#> colGraphs(1): cellcor

# Example 2: From `SingleCellExperiment` object ----
sce <- SingleCellExperiment(assays = list(counts = mat))
ge <- as(sce, "GraphExperiment")
ge
#> class: GraphExperiment 
#> dim: 200 100 
#> metadata(0):
#> assays(1): counts
#> rownames(200): gene1 gene2 ... gene199 gene200
#> rowData names(0):
#> colnames(100): cell1 cell2 ... cell99 cell100
#> colData names(0):
#> reducedDimNames(0):
#> mainExpName: NULL
#> altExpNames(0):
#> rowGraphs(0):
#> colGraphs(0):