In cbg-ethz/pareg: Pathway enrichment using a regularized regression approach

Prelude

library(tidyverse)
library(ComplexHeatmap)
library(circlize)
library(GGally)

library(pareg)
data(pathway_similarities, package = "pareg")

set.seed(42)

Introduction

Pathway similarities describe how similar two pathways are (you're welcome). For example, when interpreting pathways as gene sets, one could count how many genes are shared between two sets. Many more sophisticated methods, such as the Jaccard index, exist [@simplifyEnrichment].

pareg provides various pre-computed similarity measures (r names(pathway_similarities[[1]])) for selected pathway databases (r names(pathway_similarities)) in matrix form.

mat <- pathway_similarities$`C2@CP:KEGG`$jaccard %>%
  as_dense_sim()
mat[1:3, 1:3]

Heatmap(
  mat,
  name = "similarity",
  col = colorRamp2(c(0, 1), c("white", "black")),
  show_row_names = FALSE,
  show_column_names = FALSE
)

Comparison of similarity measures

On the Gene Ontology's Biological Process subcategory, we can observe how much pathway similarity measures can differ from each other.

df_sim <- pathway_similarities$`C5@GO:BP` %>%
  map_dfr(function(mat) {
    if (is.null(mat)) {
      return(NULL)
    }

    mat %>%
      as_dense_sim() %>%
      as.data.frame %>%
      rownames_to_column() %>%
      pivot_longer(-rowname)
  }, .id = "measure") %>%
  filter(value > 0) %>%
  pivot_wider(names_from = measure, values_from = value) %>%
  select(-rowname, -name)

ggpairs(df_sim) +
  theme_minimal()