Find consensus co-expression modules across studies

Builds a gene-gene Pearson correlation matrix on log-CPM expression in each study, combines them into a consensus (by default the element-wise minimum, as in WGCNA consensus networks, so a link must hold in every study), clusters genes hierarchically, and cuts the tree into modules. Each module is summarised by its eigengene (first principal component) in every study; eigengene differences between cases and controls are pooled across studies with a random-effects model.

Usage

mx_coexpression_modules(
  studies,
  genes = NULL,
  n_genes = 2000,
  consensus = c("min", "mean"),
  cut_height = 0.25,
  min_module_size = 20,
  meta_result = NULL
)

Arguments

studies: A named list of metaXpressStudy objects.
genes: Optional character vector of genes to cluster (e.g. meta-DEGs). Default NULL uses the n_genes most variable genes shared by all studies.
n_genes: Number of most variable genes to use when genes is NULL. Default 2000.
consensus: How to combine study correlation matrices: "min" (conservative, default) or "mean".
cut_height: Tree cut height on the distance \((1 - r)/2\). The default 0.25 groups genes whose average consensus correlation is at least about 0.5.
min_module_size: Modules smaller than this are merged into module "M0" (unassigned). Default 20.
meta_result: Optional metaXpressResult. If given, each module reports the fraction of its genes that are significant meta-DEGs (meta_padj <= 0.05) up or down.

Value

A list with

modules: data.frame of gene_id, module, kME (mean correlation with the module eigengene across studies).
module_stats: data.frame with one row per module: module, n_genes, meta_g, pvalue, padj, i_squared, per-study g_<study>, hub_genes, and optionally pct_meta_up, pct_meta_down.
eigengenes: List of per-study sample x module eigengene matrices.

Details

Like mx_cell_composition, a lib_size column in colData overrides the column sums used for log2-CPM.

References

Langfelder, P. and Horvath, S. (2007) Eigengene networks for studying the relationships between co-expression modules. BMC Systems Biology, 1, 54. doi:10.1186/1752-0509-1-54

Langfelder, P. and Horvath, S. (2008) WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics, 9, 559. doi:10.1186/1471-2105-9-559

Examples

if (FALSE) { # \dontrun{
  net <- mx_coexpression_modules(studies, n_genes = 1500,
                                 meta_result = meta_result)
  net$module_stats
} # }