Calculate random network AUPR.

calculate_random_network_AUPR.R

# Calculate random network AUPR.

## Author: Ji Huang
## Date: 2020-04-15


library(here)
library(tidyverse)
library(precrec)

# function to manipulate the edge for each network.
confirm_edge <- function(network) {
    mutate({{network}}, edge = paste0(regulator, "-", target)) %>%
        arrange(desc(weight)) %>% 
        filter(regulator %in% all_regulator) %>% 
        mutate(label = if_else(edge %in% filter_edge, "1", "0")) %>% 
        select(weight, label) %>% 
        mutate(label = as.integer(label))
}

#  function to calculate a random network AUPR.
calc_random_ntwk_aupr <- function(i) {
    set.seed(i)
    random1 <- op3 %>% mutate(weight = sample(weight, replace = F))
    random1 <- confirm_edge(random1)
    msmdat <- mmdata(random1$weight, random1$label)
    rand_aupr <- auc(evalmod(msmdat))[2,4]
    return(rand_aupr)
}


## Load OutPredict result.
op3 <- read_csv(here("result", "Ranked_list_TF_gene_best_model_Nonly_ntree1000.csv"),
                col_names = c("regulator", "target", "weight"),
                skip = 1) %>% 
    filter(regulator!= target)

## Load the positive TARGET result. Keep the 29 TFs within the network (Discard 4 TFs)
ptarget <- read_tsv(here("data", "at_33target_result.txt"))
all_regulator <- unique(op3$regulator)[unique(op3$regulator) %in% 
                                           unique(ptarget$regulator)]

## Combine TF and edges, filter 29 TFs and check whether edges are confirmed by TARGET.
ptarget <- ptarget %>% filter(regulator %in% all_regulator) %>% 
    mutate(edge = paste0(regulator, "-", target))

filter_edge <- ptarget$edge

## Repeat 10000 times for random network.
random_aupr <- map_dbl(1:10000, calc_random_ntwk_aupr)
random_aupr <- data.frame(aupr = random_aupr)
write_tsv(random_aupr, path = here("result", "ath_root_random_network_1E5rep.tsv.gz"))

## Take a quick look.
tt1 <- read_tsv(here("result", "ath_root_random_network_1E5rep.tsv.gz"))
fivenum(tt1$aupr)
plot(density(tt1$aupr))