GraphExperiment: infrastructure for high-dimensional quantitative data and associated networks in R/Bioconductor

Abstract

Networks are widely used to represent relationships among features (e.g. genes, proteins, and microbial taxa) and among observations (e.g. cells, samples, and species) in omics data, as well as their neighborhoods. The Bioconductor ecosystem provides users and developers with modular and interoperable software tools by relying on standard data structures that are reused across packages. However, Bioconductor lacks a standard data structure to store such networks together with the quantitative data from which they are inferred. Here, we present GraphExperiment, an R/Bioconductor package that provides an S4 class extending SingleCellExperiment with containers for feature- level (rowGraphs) and observation-level (colGraphs) networks stored as igraph objects. Networks are kept in sync with assays, row and column metadata, allowing the simultaneous manipulation of quantitative data and graphs. We demonstrate GraphExperiment in seven use cases covering bulk and single-cell transcriptomics, single-cell proteomics, microbiome, and comparative genomics data, with networks ranging from gene coexpression and protein co-abundance networks to cell-cell communication and phylogenetic networks. GraphExperiment is available on Bioconductor (https://bioconductor.org/packages/GraphExperiment).