perm_edit_tmp.R

setwd("/Users/benjamin/Repositories/Zuzu/work/75/1dbc4e5fcd451d8ae9a1a6ccb8aeee/")

library("tidyverse")

#print(paste0("DEBUG: ", "samplesheet.csv", " ", "1"))

samplesheet = read.csv("samplesheet.csv")
countsframe.clean = read.csv("countsframe_clean.csv", row.names = 1, check.names = FALSE)

# Permute colnames at random while preserving gene counts
set.seed(1)
permute_fun = function(x){
  for(i in 1:100){ # NB if we don't get a proper permutation after 100 runs, we give up because the sample sizes are obviously inappropriate
    # Permute and check whether any of the newly permuted rows have perfectly recaptured the original groupings
    permrow = sample(x,replace = FALSE) 
    newcols = samplesheet$phenotype[match(names(permrow), samplesheet$sample)]
    proptrue = sum(newcols == samplesheet$phenotype)/length(newcols)
    # If more than 80% of sample labels have remained the same or have flipped, re-permute everything
    if(abs(proptrue-0.5)<0.3){
      break
    }else if(i==100){
      stop("Unable to achieve a balanced permutation. Are your sample sizes large enough?")
    }
  }
  return(data.frame(t(permrow)))
}

#coerce back to df
countsframe.perm = apply(countsframe.clean, 1, permute_fun) %>% 
  bind_rows(., .id = "column_label") %>%
  column_to_rownames(var = "column_label") %>%
  `colnames<-`(colnames(countsframe.clean))

write.csv(samplesheet,"samplesheet_perm.csv", row.names = FALSE)
write.csv(countsframe.perm,"countsframe_perm.csv")