wgcna-toolbox

Perslab toolbox for Weighted Gene Co-Expression Network Analysis

Robust WGCNA pipeline (rwgnca_main_seurat3.0.R)

Dependencies:

• optparse
• parallel
• Seurat >= 3.0
• WGCNA >- 1.68
• magrittr
• Matrix
• data.table
• Biobase
• reshape
• reshape2

Quickstart

Make sure you have log-normalized expression data and metadata with cell names in the first column. Then call

time Rscript ./rwgcna_main_seurat3.0.R --pathDatExpr ./expressionData.csv.gz --pathMetadata ./metadata.csv --dirProject /my/project/dir/ --prefixData mousebrain_neurons --prefixRun 1 --dataType sc --colIdents ClusterName 

Parameter hints

Rscript ./rwgcna_main_seurat3.0.R --help

Workflow details

Subset the data and select variable genes for WGCNA

• Split cells into subsets by annotation (i.e. celltype for single-cell, tissue-type for bulk RNA-seq)
• Select genes for the analysis with the arguments featuresUse and nFeatures. Use either var.features for variable features using Seurats FindVariableFeatures function (vst), PCLoading for top loading PC genes, JackStrawPCLoading for top loading PC genes PCs found to be significant using the JackStraw test, or JackStrawPCSignif to select genes with best p-values on significant PCs identified by the JackStraw test.

Compute adjacency matrix, consensus Topological Overlap Matrix (TOM) and find gene modules

• Compute soft power for adjacency matrix
• Resample expression matrices (.66, without replacement)
• Compute consensus TOM across resampled expression matrices
• Run hclust to find clusters in the (inverse) TOM
• Run cutreeHybrid to cut out modules in the distance matrices (iterate over different sets of parameters if given)
• Merge close modules iteratively using kIMs or kMEs
• Compute fuzzy module membership for every gene-module pair, either kMEs ('fuzzy' module membership) or intramodular k (kIMs), which are average distance between a gene and each gene in the module
• Perform an additional k-means-like reclustering step, reassigning genes to modules if the kME/kIM is more than 1.2 times higher
• Filter out genes without a significant gene-module expression profile correlation using a t-test
• Make module names unique across all celltypes/tissues

Outputs

Results:

• table with columns cell type, module, gene and gene weight/centrality (kME or kIM)
• module gene weights for all genes with respect to all modules (kME / kIM)
• matrix of original expression data embedded into module space

Metadata:

• table of parameters
• table with run summary statistics
• table with per celltype subset summary statistics
• final session image with all of the above (in \RObjects subdir)

Issues / bugs

Submit an issue with

• Descriptive title summarising the issue
• What you did
• Expected behaviour
• Actual behaviour
• R version and, if relevant, package versions
• Relevant output