Derive TME meta-programs by clustering Hallmarks across NMF factors

Hierarchically clusters (Ward.D2 on Euclidean distance) Hallmark gene sets by their NES profile across NMF factors to identify recurrent transcriptional programs in the TME. Use annotate_metaprograms_TME() afterwards to label each meta-program with a Bagaev et al. (2021) MFP subtype.

Usage

derive_meta_programs(gsea_results, k = NULL, file_name = NULL, plot = TRUE)

Arguments

gsea_results

Output of compute_factor_gsea() (a list with a GSEA_results element, one table per NMF factor).

k

Integer. Number of meta-programs (clusters) to extract. If NULL (default), chosen at the elbow of the within-cluster sum of squares (the k with the largest second difference) for k = 2 to min(10, n_hallmarks - 1).

file_name

Optional character suffix for the saved heatmap (Results/TCGA_meta_programs_<file_name>.pdf).

plot

Logical. If TRUE (default), saves a clustering heatmap.

Value

A data frame with one row per meta-program and columns meta_program ("MP1", "MP2", ...) and hallmarks (comma-separated Hallmark gene set names).