Hypergraph for Dynamic Correlation

R code for methods proposed by https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-019-5787-x.

Prerequisites

The following packages are required for executing the main code file: igraph, coin, ks, dynamicTreeCut, GOstats. For the sample dataset in the below example, org.Sc.sgd.db is also needed.

Usage

Data formats

• A gene expression matrix (called "array"), seen in spellman_73_filled.bin for this example
• A list of GO gene module selection (called "GO.select"), seen in GO_select_yeast_ 0.7 0.4 1000 20 .bin for this example

Preliminary process

Suposse all relevant files are in the same folder, and the work directory has been set properly. First, load in data (as demonstration, we only random sample 1000 genes here)

load('spellman_73_filled.bin')
load("GO_select_yeast_ 0.7 0.4 1000 20 .bin")
array <- array[sample(nrow(array), 1000),]

and load the main code file

source("main.r")

NOTE: The program is running on a Windows machine by default.

If you are using other operating systems, in main.r, find and comment out the line

dyn.load("csupp.dll") 

and uncomment the line

dyn.load("csupp.so")

for Linux, or uncomment the line

dyn.load("csupp_OS/csupp.so")

for Mac OS.

Relevant files are all provided in this repository.

Unsupervised hypergraph construction

To construct module-level hyergraph, use

hp_us <- hypergraph_unsup(array, fdr=0.2, save_glist=T)

and visualize the full hypergraph and top-connected sub-hypergraph by

full_hypergraph_us <- plot_entrie_unsup(hp_us$net, hp_us$label, folds=2)
plot_top_unsup(full_hypergraph_us$g, full_hypergraph_us$elist, hp_us$label, full_hypergraph_us$folds)

(Figure1 and Figure2)

One can also conduct GO enrichment test using

enrichment <- GOen(hp_us$glist, hp_us$label)

which generates a list of enrichment results for each gene cluster.

Supervised hypergraph construction

To construct module-level hyergraph, use

hp_sp <- hypergraph_sup(array, GO.select, fdr=0.2, save_triplets=T, save_glist=T)

and visualize the full hypergraph, top-connected sub-hypergraph and module-specified sub-hypergraph by

full_hypergraph_sp <- plot_entrie_sup(hp_sp$net, hp_sp$label, GO.select, folds=10)
plot_top_sup(full_hypergraph_sp$g, full_hypergraph_sp$elist, hp_sp$label, GO.select, full_hypergraph_sp$folds)
plot_one_sup(full_hypergraph_sp$g, full_hypergraph_sp$elist, hp_sp$label, GO.select, full_hypergraph_sp$folds, 
             "GO:0006090", max_num_display = 16)

(Figure3, Figure4 and Figure5)

Suppose one is interested in the gene level hypergraph given a specific hyperedge:

GO <- names(GO.select)
module_names <- Term(GO)
which(GO == "GO:0006090")
which(GO == "GO:0045937")
which(GO == "GO:2000241")
hyperedge <- c(232, 141, 1)

plot_gene_level(hyperedge, GO, hp_sp$net, hp_sp$label, hp_sp$triplets, hp_sp$glist, folds=10)

(Figure6)