R/contingency.R

In InouyeLab/NetRep: Permutation Testing Network Module Preservation Across Datasets

### Calculate a contigency table of module overlap between datasets
### 
### @param modAssignments a list where the first element is the 
###  'moduleAssignments' vector in the discovery dataset, and the second element
###  is the 'moduleAssignments vector in the test dataset.
### @param mods the 'modules' vector for the discovery dataset.
### @param tiNodelist a vector of node IDs in the test dataset.
### 
### @return 
###  A list containing a contigency table, a vector of the proportion of
###  nodes present in the test dataset for each module, a vector containing
###  the number of nodes present in the test dataset for each module, a vector
###  of the node names present in both the discovery and test datasets, a vector
###  of modules that are both requested and have nodes present in the test 
###  dataset, and the \code{modAssignments} vector containing only nodes present
###  in the test dataset.
### 
### @keywords internal
contingencyTable <- function(modAssignments, mods, tiNodelist) {
  # To simplify later function calls, we need to get a vector of module
  # assignments only for (a) modules of interest and (b) the variables
  # present in both datasets for those modules.
  overlapVars <- intersect(names(modAssignments[[1]]), tiNodelist)
  overlapAssignments <- modAssignments[[1]][overlapVars]
  
  overlapAssignments <- overlapAssignments %sub_in% mods
  overlapModules <- unique(overlapAssignments)
  overlapModules <- overlapModules[orderAsNumeric(overlapModules)]
  
  # How many variables are present in the test dataset for the modules 
  # of interest?
  varsPres <- table(overlapAssignments)
  
  modulesWithNoOverlap <- mods %sub_nin% overlapAssignments
  varsPres <- c(varsPres, rep(0, length(modulesWithNoOverlap)))
  names(varsPres)[names(varsPres) == ""] <- modulesWithNoOverlap
  varsPres <- varsPres[orderAsNumeric(names(varsPres))]
  
  if (any(varsPres == 0)) {
    noNodes <- names(varsPres[varsPres == 0])
    warning(
      "None of the nodes in module(s) ", 
      paste(paste0('"', noNodes, '"'), collapse=", "),
      " are present in the test dataset",
      immediate. = TRUE
    )
  }
  
  # What proportion?
  moduleSizes <- table(modAssignments[[1]])
  moduleSizes <- moduleSizes[names(moduleSizes) %sub_in% mods]
  propVarsPres <- varsPres / moduleSizes[names(varsPres)]

  # Calculate some basic cross-tabulation statistics so we can assess 
  # which modules in both datasets map to each other, if module
  # detection has also been performed for the test network
  contingency <- NULL
  if (!is.null(modAssignments[[2]])) {
    # Get total number of nodes from each discovery subset in each test subset 
    contingency <- table(
      modAssignments[[1]][overlapVars], 
      modAssignments[[2]][overlapVars]
    )
    
    ## Add in the sizes of each module, and the number of variables present in
    ## the other dataset
    
    # Get the module sizes irrespective of dataset overlap
    discSizes <- table(modAssignments[[1]])
    testSizes <- table(modAssignments[[2]])
    
    # The dummy row / column handles modules with no variables present
    discInfo <- cbind(discSizes, 0)
    testInfo <- rbind(testSizes, 0)
    colnames(discInfo) <- c("size", "present")
    rownames(testInfo) <- c("size", "present")

    # Add the number of variables present for each module in the other dataset
    discPresent <- table(modAssignments[[1]][overlapVars])
    testPresent <- table(modAssignments[[2]][overlapVars])
    discInfo[names(discPresent), "present"] <- discPresent
    testInfo["present", names(testPresent)] <- testPresent
    
    # Add missing modules to the contigency table
    discMiss <- rownames(discInfo) %sub_nin% rownames(contingency)
    testMiss <- colnames(testInfo) %sub_nin% colnames(contingency)
    discMissMat <- matrix(0, nrow=length(discMiss), ncol=ncol(contingency),
                          dimnames=list(discMiss, colnames(contingency)))
    contingency <- rbind(contingency, discMissMat)
    testMissMat <- matrix(0, ncol=length(testMiss), nrow=nrow(contingency),
                          dimnames=list(rownames(contingency), testMiss))
    contingency <- cbind(contingency, testMissMat)
    
    # Order and filter contingency table prior to adding information
    contingency <- contingency[mods,, drop=FALSE]
    contingency <- contingency[orderAsNumeric(rownames(contingency)), 
                               orderAsNumeric(colnames(contingency)),
                               drop=FALSE]
    
    # Make sure the info tables are in the same order so we can add them 
    # correctly
    discInfo <- discInfo[rownames(contingency),, drop=FALSE]
    testInfo <- testInfo[, colnames(contingency), drop=FALSE]
    
    # Create the empty upper left square
    na_mat <- matrix(NA, nrow=2, ncol=2) 
    dimnames(na_mat) <- list(c("size", "present"), c("size", "present"))
    
    # combine
    contingency <- cbind(discInfo, contingency)
    contingency <- rbind(cbind(na_mat, testInfo), contingency)
  }
  
  return(list(
    contingency=contingency, propVarsPres=propVarsPres, overlapVars=overlapVars,
    varsPres=varsPres, overlapModules=overlapModules,
    overlapAssignments=overlapAssignments
  ))
}