Network-Based Clustering of Pan-Cancer Data Accounting for Clinical Covariates

This repository contains the code to reproduce the results of the NeurIPS 2022 LMRL workshop paper "Network-Based Clustering of Pan-Cancer Data Accounting for Clinical Covariates".

Installation

In order to install the package, it suffices to launch R CMD INSTALL path/to/graphClust from a terminal, or make install from within the package source folder.

Being hosted on GitHub, it is possible to use the install_github tool from an R session:

if (!requireNamespace("BiocManager", quietly = TRUE)) install.packages("BiocManager")
BiocManager::install(c("Rgraphviz", "RBGL"))

library("devtools")
install_github("cbg-ethz/graphClust_NeurIPS")

graphClust requires R >= 3.5, and depends on pcalg, reshape2, BiDAG (>= 2.0.2), RBGL, clue and grDevices.

Simulations

Figure 2 can be reproduced by running the script simulations/figure_2-simulation.R. Analogously, Figure 4 in the appendix can be reproduced by running the script simulations/figure_4-simulation.R. The simulations can be modified and executed in the simulations/cluster-scripts folder.

Pan-Cancer Data

Figure 3 can be reproduced by runnign the script tcga_analysis/figure_3-km_plot.R. The results of Table 1 can be reproduced by runnign the script tcga_analysis/table_1-cox_analysis.R. A reproducability analysis for a range of different seeds can be found in tcga_analysis/reproducability_different_seeds. The hyperparameters of the cluster algorithms can be modified and executed in the tcga_analysis/clustering folder folder.

Example

library(graphClust)

# Simulate binary data from 3 clusters
k_clust <- 3
ss <- c(400, 500, 600) # samples in each cluster
simulation_data <- sampleData(k_clust = k_clust, n_vars = 20, n_samples = ss)
sampled_data <- simulation_data$sampled_data

# Network-based clustering
cluster_res <- get_clusters(sampled_data, k_clust = k_clust)

# Calculate the ARI 
library(mclust)
adjustedRandIndex(simulation_data$cluster_membership, cluster_res_t$clustermembership)

# Visualize the networks
library(ggplot2)
library(ggraph)
library(igraph)
library(ggpubr)

graphClust::plot_clusters(cluster_res_t)

# Visualize a single network
my_graph <- igraph::graph_from_adjacency_matrix(cluster_res_t$DAGs[[1]], mode="directed")
graphClust::nice_DAG_plot(my_graph)