R/getFeatureScores.R

In BaderLab/netDx: Network-based patient classifier

#' Compile network scores into a matrix
#'
#' @details Given network scores over a set of train/test splits, compiles 
#' these into a matrix for downstream analysis. See the section on 
#' 'Output Files'
#' @param inDir (char/list) directory containing directories with all split 
#' info or list of all CV score files.
#' if inDir is a single directory then the expected format for CV score files 
#' is <inDir>/rngX/predClassX/GM_results/predClassX_pathway_CV_score.txt'
#' if inDir is a list, it should have one key per class. The value should be 
#' the corresponding set of filenames for pathway_CV_score.txt
#' @param predClasses (char) possible STATUS for patients
#' @param getFullCons (logical) if TRUE, does not remove rows with NA.
#' Recommended only when the number of input features is extensively 
#' pruned by first-pass feature selection.
#' @return (list) one key per patient class. Value is matrix of network
#' scores across all train/test splits. Each score is the output of
#' the inner fold of CV.
#' @examples
#' inDir <- system.file("extdata","example_output",package="netDx")
#' netScores <- getFeatureScores(inDir, predClasses = c('LumA','notLumA'))
#' @importFrom methods is
#' @export
getFeatureScores <- function(inDir, predClasses, getFullCons = TRUE) {
    if (missing(inDir)) 
        stop("inDir not provided")
    if (missing(predClasses)) 
        stop("predClasses missing; please specify classes")
    
    out <- list()
    for (gp in predClasses) {
        message(sprintf("%s\n", gp))
        
        if (is(inDir, "character")) {
            message("\tSingle directory provided, retrieving CV score files\n")
            rngDirs <- dir(path = inDir, pattern = "^rng")
            fList <- paste(inDir,rngDirs,gp,"GM_results",
			sprintf("%s_pathway_CV_score.txt",gp),
			sep=getFileSep()) 
        } else {
            message("\tList of filenames provided\n")
            fList <- inDir[[gp]]
        }
        
        message(sprintf("Got %i iterations", length(fList)))
        netColl <- list()
        
        for (scoreFile in fList) {
            tmp <- read.delim(scoreFile, sep = "\t", header = TRUE, 
							as.is = TRUE)
            colnames(tmp)[1] <- "PATHWAY_NAME"
            netColl[[scoreFile]] <- tmp
        }
        spos <- gregexpr("\\/", fList)
        # get the name of the iteration (rngX) assuming directory structure
        # rngX/<class>/GM_results>/pathway_CV_score.txt
        fNames <- lapply(seq_len(length(spos)), function(x) {
            n <- length(spos[[x]])
            y <- substr(fList[x], spos[[x]][n - 3] + 1, spos[[x]][n - 2] - 1)
            y
        })
        fNames <- unlist(fNames)
        names(netColl) <- fNames
        
        # filter for nets meeting cutoff criteria
        message("* Computing consensus\n")
        
        cons <- getNetConsensus(netColl)
        x1 <- nrow(cons)
        na_sum <- rowSums(is.na(cons))
        full_cons <- cons
        cons <- cons[which(na_sum < 1), ]
        
        if (getFullCons) 
            out[[gp]] <- full_cons else out[[gp]] <- cons
    }
    return(out)
}