#' Compute TF network and add it to hummus object
#'
#' Compute a protein-protein interaction layer from Omnipath request that will represent tf cooperativity.
#' This network is the top-layer of HuMMuS multilayer.
#'
#' @param hummus (Hummus_Object) - Hummus object
#' @param organism (integer) - Specie identifier from Omnipath to fetch
#' specific interactions
#' @param tfs vector(character) - List of tfs consider. If NA, tfs are extracted
#' from the hummus object with get_tfs function.
#' @param gene_assay (character) - Name of the assay to get tfs from if tfs is
#' not provided. If NULL, all TFs with motifs in the hummus object are used.
#' @param method (character) - Method used to infer network edges.
#' * \code{'Omnipath'} - Use Omnipath to infer tf-tf networks.
#' * \code{'NULL'} - A fake connected network is computed.
#' * \code{'Other method'} - TO DO.
#' @param store_network (bool) - Save the network directly (\code{TRUE},
#' default) or return without saving on disk (\code{FALSE}).
#' @param output_file (character) - Name of the output_file
#' (if store_network == \code{TRUE}).
#' @param source_target ('AND'|'OR') - Fetch only the interactions involving
#' two considered tfs (\code{'AND', default}), or one considered tfs and any
#' other element (\code{'OR'})
#' @param multiplex_name (character) - Name of the multiplex to add the network
#' to. Default is \code{'TF'}.
#' @param tf_network_name (character) - Name of the network in the multiplex to
#' add the network to. Default is \code{'TF_network'}.
#' @param verbose (integer) - Display function messages. Set to 0 for no message
#' displayed, >= 1 for more details.
#'
#' @return (Hummus_Object) - Return hummus object with the new network added.
#' @export
#'
#' @examples hummus <- compute_tf_network(hummus,
#' gene_assay = "RNA",
#' verbose = 1)
compute_tf_network <- function(
hummus, # Hummus object
organism = 9606, # Human by default
tfs = NA, # List of tfs considered.
gene_assay = NULL, # Name of the assay to get tfs from
# if tfs is not provided
method = NULL, # Method used to infer network edges.
# * 'Omnipath' - Use Omnipath to infer tf-tf networks.
# * 'NULL' - A fake connected network is computed.
# * 'Other method' - TO DO.
store_network = FALSE, # Save the network on disk (TRUE, default)
output_file = NULL, # Name of the output_file (if store_network == TRUE)
source_target = "AND", # 'AND' | 'OR'
multiplex_name = "TF", # Name of the multiplex to add the network to
tf_network_name = "TF_network", # Name of the network in the multiplex
verbose = 1
) {
a <- Sys.time()
# Check if method is implemented
if (is.null(method)) {
tf_network <- run_tf_null_wrapper(
hummus = hummus,
organism = organism,
tfs = tfs,
gene_assay = gene_assay,
verbose)
} else if (method == "Omnipath") {
if (!requireNamespace("OmnipathR", quietly = TRUE)) {
stop("Please install Omnipath.\n",
"github.com/saezlab/OmnipathR")
} else {
# infer network with cicero
tf_network <- run_omnipath_wrapper(
hummus = hummus,
organism = organism,
tfs = tfs,
gene_assay = gene_assay,
source_target = source_target,
verbose = verbose)
}
} else {
stop(cat("Method not implemented yet, choose between Omnipath and NULL..",
"that's it for now.\n But you can always compute the network",
"independently and add it to the hummus object manually !"))
}
if (verbose > 0) {
cat("TF network construction time:", Sys.time() - a)
}
# Save gene network
store_network(network = tf_network,
store_network = store_network,
output_file = output_file,
verbose = verbose)
# Add network to hummus object
hummus <- add_network(hummus,
multiplex_name = multiplex_name,
network = tf_network,
network_name = tf_network_name,
weighted = FALSE, # PPI could be weighted,
# could be added later
directed = FALSE, # PPI are not directed
verbose = verbose)
return(hummus)
}
#' Compute gene netwok from scRNA-seq data
#'
#' This function will create a network from rna data (or in theory any data
#' wtih genes as features).
#' Different method should be implemented at some point (any suggestion is welcomed ! :) ),
#' for now Genie3 is still the reference and only method available
#'
#' Method descriptions :
#' 1. Genie3
#' Use tree random forest to infer regulatory networks :
#' https://bioconductor.org/packages/release/bioc/html/GENIE3.html
#'
#' @param hummus (Hummus_Object) - Hummus object
#' @param gene_assay (character) - Name of the assay containing the gene
#' expression data.
#' @param tfs vector(character) - List of tfs considered. If NULL, all TFs with
#' motifs in the hummus object are used.
#' @param method (character) - Method used to infer network edges.
#' * \code{'Genie3'} - Use tree random forest to infer regulatory networks.
#' * \code{'Other method'} - TO DO.
#' @param multiplex_name (character) - Name of the multiplex to add the network
#' to. Default is \code{'RNA'}.
#' @param network_name (character) - Name of the network in the multiplex to
#' add the network to. Default is \code{'RNA_network'}.
#' @param store_network (bool) - Save the network directly (\code{TRUE},
#' default) or return without saving on disk (\code{FALSE}).
#' @param output_file (character) - Name of the output_file
#' (if store_network == \code{TRUE}).
#' @param threshold (interger, default 0) - Minimal threshold
#' to select tf-gene edges.
#' @param number_cores (interger, default 1) - Number of thread that should be
#' used for the parallelizable methods.
#' @param verbose (integer) - Display function messages. Set to 0 for no
#' message displayed, >= 1 for more details.
#'
#' @return (data.frame) - Return list of network interactions between genes
#' @export
#'
#' @examples hummus <- compute_gene_network(
#' hummus,
#' gene_assay = "RNA",
#' method = "GENIE3",
#' verbose = 1,
#' number_cores = 8,
#' store_network = FALSE)
#'
compute_gene_network <- function(
hummus,
gene_assay = "RNA",
tfs = NULL,
method = "GENIE3",
multiplex_name = NULL,
network_name = NULL,
store_network = FALSE,
output_file = NULL,
threshold = 0.0,
number_cores = 1,
verbose = 1
) {
a <- Sys.time()
# Check if method is implemented
if (method == "GENIE3") {
if (verbose > 0) {
cat("Computing gene network with ", method, " ...\n")
}
# Get tfs list
if (verbose > 0 && is.null(tfs)) {
cat("\tNo TFs list provided, fetching from hummus object...\n")
tfs <- get_tfs(hummus = hummus,
assay = gene_assay,
store_tfs = FALSE,
output_file = NULL,
verbose = verbose)
}
# infer network
weightMat <- GENIE3::GENIE3(
as.matrix(hummus@assays[[gene_assay]]$counts),
regulators = tfs,
nCores = number_cores)
# get edge list
linkList <- GENIE3::getLinkList(weightMat)
gene_network <- linkList[which(linkList$weight > threshold), ]
# TODO : add other methods
} else {
stop(cat("Method not implemented yet, choose between GENIE3 and..",
"that's it for now.\n but you can always compute the network",
"independently and add it to the hummus object."))
}
if (verbose > 0) {
cat("\tGene network construction time:", Sys.time() - a, "\n")
}
# Save gene network
store_network(network = gene_network,
store_network = store_network,
output_file = output_file,
verbose = verbose)
# If no multiplex name provided, use assay name
if (is.null(multiplex_name)) {
multiplex_name <- gene_assay
}
# If no network name provided, use method name + assay name
if (is.null(network_name)) {
network_name <- paste(multiplex_name, method, sep = "_")
}
# Add network to hummus object
hummus <- add_network(hummus,
multiplex_name = multiplex_name,
network = gene_network,
network_name = network_name,
weighted = TRUE,
directed = FALSE,
verbose = verbose)
# Return hummus object
return(hummus)
}
#' Compute peak network from scATAC-seq data
#'
#' This function will create a network from atac data (or in theory any data
#' wtih peaks coordinates as features).
#' Different method should be implemented at some point (e.g. RENIN),
#' for now Cicero is still the reference and only method available
#'
#' Method descriptions :
#' 1. Cicero
#' Use patial corelation between peaks that are in a given window (e.g. :
#' less distant than 500K base pairs)
#'
#' @param hummus (Hummus_Object) - Hummus object
#' @param atac_assay (character) - Name of the assay containing the atac
#' peaks data.
#' @param genome (BSgenome) - Genome used to compute the distance between peaks.
#' @param method (character) - Method used to infer network edges.
#' * \code{'cicero'} - Use cicero to infer regulatory networks.
#' * \code{'Other method'} - TO DO.
#' @param multiplex_name (character) - Name of the multiplex to add the network
#' to. Default is \code{'peaks'}.
#' @param network_name (character) - Name of the network in the multiplex to
#' add the network to. Default is \code{'peak_network'}.
#' @param store_network (bool) - Save the network directly (\code{TRUE},
#' default) or return without saving on disk (\code{FALSE}).
#' @param output_file (character) - Name of the output_file
#' (if store_network == \code{TRUE}).
#' @param threshold (interger, default 0) - Minimal threshold to select tf-gene
#' edges.
#' @param number_cells_per_clusters (integer) - Number of cells grouped by
#' territory to define pseudocells
#' @param sample_num (integer | Cicero) - Number of pseudocells to sample from
#' each territory. Default is 100.
#' @param seed (integer | Cicero) - Seed used to sample pseudocells. Default is
#' 2025
#' @param verbose (integer) - Display function messages. Set to 0 for no
#' message displayed, >= 1 for more details.
#' @param window (integer) - Size of window to consider potential
#' cis-regulatory cooperations between peaks. Default is 500K base pairs.
#' @param reduction_method (character | Cicero) - Method used to reduce dimensionality
#' of the data to identify territories. Default is \code{'UMAP'}.
#'
#' @return (data.frame) - Return list of network interactions between peaks
#' @export
#'
#' @examples hummus <- compute_atac_peak_network(hummus)
#'
compute_atac_peak_network <- function(
hummus,
atac_assay = "peaks",
genome = BSgenome.Hsapiens.UCSC.hg38::BSgenome.Hsapiens.UCSC.hg38,
method = "cicero",
multiplex_name = NULL,
network_name = NULL,
store_network = FALSE,
output_file = NULL,
threshold = 0.0,
number_cells_per_clusters = 50,
sample_num = 100,
seed = 2025,
verbose = 1,
window = 5e+05,
reduction_method = "UMAP") {
a <- Sys.time()
# Check if method is implemented
if (method == "cicero") {
if (!requireNamespace("cicero", quietly = TRUE)) {
stop("Please install cicero.\n",
"https://cole-trapnell-lab.github.io/cicero-release/docs_m3/")
} else {
# infer network with cicero
atac_peak_network <- run_cicero_wrapper(
hummus,
atac_assay,
genome,
window,
number_cells_per_clusters,
sample_num,
seed,
verbose,
threshold,
reduction_method)
}
} else {
stop(cat("Method not implemented yet, choose between Cicero and..",
"that's it for now.\n but you can always compute the network",
"independently and add it to the hummus object manually."))
}
if (verbose > 0) {
cat("Peak network construction time:", Sys.time() - a)
}
# Save peak network
store_network(network = atac_peak_network,
store_network = store_network,
output_file = output_file,
verbose = verbose)
# If no multiplex name provided, use assay name
if (is.null(multiplex_name)) {
multiplex_name <- atac_assay
}
# If no network name provided, use method name + assay name
if (is.null(network_name)) {
network_name <- paste0("peak_network_", method)
}
# Add network to hummus object
hummus <- add_network(
object = hummus,
network = atac_peak_network,
network_name = network_name,
multiplex_name = multiplex_name,
weighted = TRUE,
directed = FALSE,
verbose = verbose)
}
Datasets
Models
