GitHub - jackgisby/ReducedExperiment: Containers and tools for dimensionally-reduced -omics data (original) (raw)

ReducedExperiment

ReducedExperiment provides containers for storing and manipulating dimensionally-reduced assay data. The ReducedExperiment classes allow users to simultaneously manipulate their original dataset and their decomposed data, in addition to other method-specific outputs like pathway analysis. Implements utilities and specialised classes for the application of stabilised independent component analysis (sICA) and weighted gene correlation network analysis (WGCNA).

Installation

Get the latest stable R release fromCRAN. Then install ReducedExperimentfromBioconductorusing the following code:

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

BiocManager::install("ReducedExperiment")

Alternatively, the development version of ReducedExperiment is available fromBioconductoror GitHub with:

BiocManager::install("ReducedExperiment", version = "devel")

devtools::install_github("jackgisby/ReducedExperiment")

The development version of the package is also available as a container onDockerHub.

Usage

ReducedExperiment objects are derived from SummarizedExperimentobjects, with additional slots designed to store and manipulate the outputs of common dimensionality reduction techniques.

As an example, the SummarizedExperiment described below contains gene expression data from individuals with COVID-19. It contains the following slots:

• assays - A features by samples matrix containing the expression data.
• colData - Contains a row for each sample containing phenotype data.
• rowData - Contains a row for each feature containing gene IDs.

The SummarizedExperiment objects are convenient because, when we slice the rows or columns of the expression matrix, the metadata for the rows and columns are sliced accordingly.

library("SummarizedExperiment")

se <- readRDS(system.file( "extdata", "wave1.rds", package = "ReducedExperiment" ))

se #> class: SummarizedExperiment #> dim: 500 83 #> metadata(0): #> assays(1): normal #> rownames(500): ENSG00000004799 ENSG00000007038 ... ENSG00000287935 #> ENSG00000288049 #> rowData names(2): ensembl_id gene_id #> colnames(83): C37_positive_9 C48_positive_4 ... C85_negative #> C89_negative #> colData names(8): sample_id individual_id ... case_control #> time_from_first_x

The SummarizedExperiment has two dimensions, representing the features (2,184) and samples (234).

We can perform a factor analysis on these data, the result of which is a set of reduced components and feature loadings.

library("ReducedExperiment")

fe <- estimateFactors(se, nc = 35) fe #> class: FactorisedExperiment #> dim: 500 83 35 #> metadata(0): #> assays(2): normal transformed #> rownames(500): ENSG00000004799 ENSG00000007038 ... ENSG00000287935 #> ENSG00000288049 #> rowData names(2): ensembl_id gene_id #> colnames(83): C37_positive_9 C48_positive_4 ... C85_negative #> C89_negative #> colData names(8): sample_id individual_id ... case_control #> time_from_first_x #> 35 components

This FactorisedExperiment object has an additional dimension representing the 35 factors. It also has additional slots, including:

• reduced - A samples by factors matrix containing the dimensionally-reduced data.
• loadings - Contains a features by factors matrix containing the loadings.

The ReducedExperiment objects allow users to simultaneously slice and modify the assays, rowData, colData, reduced and loadingsmatrices. Here, we provided a SummarizedExperiment object toestimateFactors, but we could just have easily provided a simple expression matrix.

Alternatively, you may have already applied dimensionality reduction to your data and simply wish to package it into a ReducedExperimentcontainer. For instance, below we apply principal components analysis, and construct a FactorisedExperiment object from the results.

prcomp_res <- stats::prcomp(t(assay(se)), center = TRUE, scale. = TRUE)

fe_prcomp <- FactorisedExperiment( se, reduced = prcomp_res$x, loadings = prcomp_res$rotation, stability = prcomp_res$sdev, center = prcomp_res$center, scale = prcomp_res$scale )

fe_prcomp #> class: FactorisedExperiment #> dim: 500 83 83 #> metadata(0): #> assays(1): '' #> rownames(500): ENSG00000004799 ENSG00000007038 ... ENSG00000287935 #> ENSG00000288049 #> rowData names(0): #> colnames(83): C37_positive_9 C48_positive_4 ... C85_negative #> C89_negative #> colData names(0): #> 83 components

Functionality

The package currently provides three types of container:

• ReducedExperiment - A basic container that can store dimensionally-reduced components.
• FactorisedExperiment - A container based on ReducedExperimentdesigned for working with the results of factor analysis. It can contain feature loadings and factor stability values.
• ModularExperiment - A container based on ReducedExperimentdesigned for working with modules of features (usually genes), as is produced by the popular Weighted Gene Correlation Network Analysis (WGCNA) approach. It contains the mapping of features to modules

Various tools are provided by the package for applying dimensionality reduction and manipulating their results. These include:

• Workflows for applying independent component analysis (ICA) and WGCNA. We additionally developed an R implementation of the stabilised ICA algorithm.
• Methods for applying pathway enrichment analysis to factors and modules.
• Functions for identifying associations between factors/modules and sample-level variables.
• Methods for applying identified factors or modules to new datasets.
• Other approach-specific plots and utilities, such as factor stability plots and module preservation plots.

Many of these are demonstrated in more detail in the package’s vignette.

The containers implemented in ReducedExperiment are designed to be extensible. We encourage the development of children classes with additional, or alternative, slots and methods.

Citation

Below is the citation output from using citation('ReducedExperiment')in R.

print(citation("ReducedExperiment"), bibtex = TRUE) #> To cite package 'ReducedExperiment' in publications use: #> #> Gisby JS, Barnes MR (2025). ReducedExperiment: Containers and tools #> for dimensionally-reduced -omics data. #> doi:10.18129/B9.bioc.ReducedExperiment #> https://doi.org/10.18129/B9.bioc.ReducedExperiment, v0.99.3, #> http://www.bioconductor.org/packages/ReducedExperiment. #> #> A BibTeX entry for LaTeX users is #> #> @Manual{, #> title = {ReducedExperiment: Containers and tools for dimensionally-reduced -omics data}, #> author = {Jack S. Gisby and Michael R. Barnes}, #> year = {2025}, #> url = {http://www.bioconductor.org/packages/ReducedExperiment}, #> note = {v0.99.3}, #> doi = {10.18129/B9.bioc.ReducedExperiment}, #> }

The ReducedExperiment package relies on many software packages and development tools. The packages used are listed in the vignette and relevant papers are cited.