Introduction
The ReactomeGSA package is a client to the web-based Reactome Analysis System. Essentially, it performs a gene set analysis using the latest version of the Reactome pathway database as a backend.
This vignette shows how the ReactomeGSA package can be used to perform a pathway analysis of cell clusters in single-cell RNA-sequencing data.
Citation
To cite this package, use
Griss J. ReactomeGSA, https://github.com/reactome/ReactomeGSA (2019)
Installation
The ReactomeGSA package can be directly installed from Bioconductor:
if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
if (!require(ReactomeGSA))
BiocManager::install("ReactomeGSA")
#> Loading required package: ReactomeGSA
# install the ReactomeGSA.data package for the example data
if (!require(ReactomeGSA))
BiocManager::install("ReactomeGSA.data")
For more information, see https://bioconductor.org/install/.
Example data
As an example we load single-cell RNA-sequencing data of B cells extracted from the dataset published by Jerby-Arnon et al. (Cell, 2018).
Note: This is not a complete Seurat object. To decrease the size, the object only contains gene expression values and cluster annotations.
library(ReactomeGSA.data)
#> Loading required package: limma
#> Loading required package: edgeR
#> Loading required package: Seurat
data(jerby_b_cells)
jerby_b_cells
#> An object of class Seurat
#> 23686 features across 920 samples within 1 assay
#> Active assay: RNA (23686 features)
Pathway analysis of cell clusters
The pathway analysis is at the very end of a scRNA-seq workflow. This means, that any Q/C was already performed, the data was normalized and cells were already clustered.
The ReactomeGSA package can now be used to get pathway-level expression values for every cell cluster. This is achieved by calculating the mean gene expression for every cluster and then submitting this data to a gene set variation analysis.
All of this is wrapped in the single analyse_sc_clusters function.
library(ReactomeGSA)
gsva_result <- analyse_sc_clusters(jerby_b_cells, verbose = TRUE)
#> Calculating average cluster expression...
#> Converting expression data to string... (This may take a moment)
#> Conversion complete
#> Submitting request to Reactome API...
#> Compressing request data...
#> Reactome Analysis submitted succesfully
#> Converting dataset Seurat...
#> Mapping identifiers...
#> Performing gene set analysis using ssGSEA
#> Analysing dataset 'Seurat' using ssGSEA
#> Retrieving result...
The resulting object is a standard ReactomeAnalysisResult object.
gsva_result
#> ReactomeAnalysisResult object
#> Reactome Release: 72
#> Results:
#> - Seurat:
#> 1720 pathways
#> 12305 fold changes for genes
#> No Reactome visualizations available
#> ReactomeAnalysisResult
pathways returns the pathway-level expression values per cell cluster:
pathway_expression <- pathways(gsva_result)
# simplify the column names by removing the default dataset identifier
colnames(pathway_expression) <- gsub("\\.Seurat", "", colnames(pathway_expression))
pathway_expression[1:3,]
#> Name Cluster.1 Cluster.10 Cluster.11
#> R-HSA-1059683 Interleukin-6 signaling 0.09545353 0.07958761 0.1330523
#> R-HSA-109606 Intrinsic Pathway for Apoptosis 0.10809265 0.10399420 0.1169451
#> R-HSA-109703 PKB-mediated events 0.17906778 0.11160170 0.1198267
#> Cluster.12 Cluster.13 Cluster.2 Cluster.3 Cluster.4 Cluster.5
#> R-HSA-1059683 0.09148179 0.09802880 0.1029586 0.09398238 0.1051934 0.09402404
#> R-HSA-109606 0.11811004 0.13732430 0.1051371 0.10962570 0.1131350 0.10520629
#> R-HSA-109703 0.14712571 0.09568951 0.1217289 0.12551990 0.1112811 0.10844562
#> Cluster.6 Cluster.7 Cluster.8 Cluster.9
#> R-HSA-1059683 0.08332742 0.1021174 0.1265275 0.09876770
#> R-HSA-109606 0.10518480 0.1160139 0.1196084 0.11431412
#> R-HSA-109703 0.17851610 0.1699756 0.1689482 0.05316363
A simple approach to find the most relevant pathways is to assess the maximum difference in expression for every pathway:
# find the maximum differently expressed pathway
max_difference <- do.call(rbind, apply(pathway_expression, 1, function(row) {
values <- as.numeric(row[2:length(row)])
return(data.frame(name = row[1], min = min(values), max = max(values)))
}))
max_difference$diff <- max_difference$max - max_difference$min
# sort based on the difference
max_difference <- max_difference[order(max_difference$diff, decreasing = T), ]
head(max_difference)
#> name min max
#> R-HSA-389542 NADPH regeneration -0.4229458 0.4292007
#> R-HSA-8964540 Alanine metabolism -0.5051647 0.2773550
#> R-HSA-140180 COX reactions -0.4743840 0.2573558
#> R-HSA-5263617 Metabolism of ingested MeSeO2H into MeSeH -0.1684564 0.4948353
#> R-HSA-9636003 NEIL3-mediated resolution of ICLs -0.4970992 0.1125057
#> R-HSA-3248023 Regulation by TREX1 -0.0969508 0.4516689
#> diff
#> R-HSA-389542 0.8521465
#> R-HSA-8964540 0.7825197
#> R-HSA-140180 0.7317398
#> R-HSA-5263617 0.6632917
#> R-HSA-9636003 0.6096049
#> R-HSA-3248023 0.5486197
Plotting the results
The ReactomeGSA package contains two functions to visualize these pathway results. The first simply plots the expression for a selected pathway:
plot_gsva_pathway(gsva_result, pathway_id = rownames(max_difference)[1])
For a better overview, the expression of multiple pathways can be shown as a heatmap using gplots heatmap.2 function:
# Additional parameters are directly passed to gplots heatmap.2 function
plot_gsva_heatmap(gsva_result, max_pathways = 15, margins = c(6,20))
The plot_gsva_heatmap function can also be used to only display specific pahtways:
# limit to selected B cell related pathways
relevant_pathways <- c("R-HSA-983170", "R-HSA-388841", "R-HSA-2132295", "R-HSA-983705", "R-HSA-5690714")
plot_gsva_heatmap(gsva_result,
pathway_ids = relevant_pathways, # limit to these pathways
margins = c(6,30), # adapt the figure margins in heatmap.2
dendrogram = "col", # only plot column dendrogram
scale = "row", # scale for each pathway
key = FALSE, # don't display the color key
lwid=c(0.1,4)) # remove the white space on the left
This analysis shows us that cluster 8 has a marked up-regulation of B Cell receptor signalling, which is linked to a co-stimulation of the CD28 family. Additionally, there is a gradient among the cluster with respect to genes releated to antigen presentation.
Therefore, we are able to further classify the observed B cell subtypes based on their pathway activity.
Pathway-level PCA
The pathway-level expression analysis can also be used to run a Principal Component Analysis on the samples. This is simplified through the function plot_gsva_pca:
plot_gsva_pca(gsva_result)
In this analysis, cluster 11 is a clear outlier from the other B cell subtypes and therefore might be prioritised for further evaluation.