scstruc

The package for analysing the gene regulatory network based on Bayesian network structure of single-cell transcriptomics data. The function works primarily with SingleCellExperiment. Multiple algorithms tailored for single-cell transcriptomics data are prepared. The inferred networks are validated based on the causal relationships between genes.

Installation

Using devtools:

devtools::install_github("noriakis/scstruc")

Documentation

The documentation is available here.

Examples

Based on SingleCellExperiment, the network is inferred and plotted.

library(scran)
library(scstruc)
library(ggraph)
library(bnlearn)
sce <- mockSCE()
sce <- logNormCounts(sce)
included_genes <- sample(row.names(sce), 100)
gs <- scstruc(sce, included_genes, changeSymbol=FALSE)
plotNet(gs$net, gs$data, showText=FALSE)

If you start from a Seurat object, convert it to SingleCellExperiment before running scstruc.

library(Seurat)
library(SeuratObject)
library(scstruc)

seu <- NormalizeData(seu)
sce <- as.SingleCellExperiment(seu)
included_genes <- sample(row.names(sce), 100)
gs <- scstruc(sce, included_genes, changeSymbol=FALSE)

Using bootstrapping, the averaged network is obtained. This time, L1MB algorithm with the selection of neighbors based on BIC is used.

library(glmnet)
gs2 <- scstruc(sce, included_genes, algorithm="glmnet_BIC", boot=TRUE,
               changeSymbol=FALSE, R=20)
#> Bootstrapping specified
plotAVN(gs2$net, sizeRange=c(1,3))