AMR/R/guess_ab_col.R

204 lines
8.4 KiB
R
Executable File

# ==================================================================== #
# TITLE #
# Antimicrobial Resistance (AMR) Analysis #
# #
# SOURCE #
# https://gitlab.com/msberends/AMR #
# #
# LICENCE #
# (c) 2019 Berends MS (m.s.berends@umcg.nl), Luz CF (c.f.luz@umcg.nl) #
# #
# This R package is free software; you can freely use and distribute #
# it for both personal and commercial purposes under the terms of the #
# GNU General Public License version 2.0 (GNU GPL-2), as published by #
# the Free Software Foundation. #
# #
# This R package was created for academic research and was publicly #
# released in the hope that it will be useful, but it comes WITHOUT #
# ANY WARRANTY OR LIABILITY. #
# Visit our website for more info: https://msberends.gitlab.io/AMR. #
# ==================================================================== #
#' Guess antibiotic column
#'
#' This tries to find a column name in a data set based on information from the \code{\link{antibiotics}} data set. Also supports WHONET abbreviations.
#' @param x a \code{data.frame}
#' @param search_string a text to search \code{x} for
#' @param verbose a logical to indicate whether additional info should be printed
#' @details You can look for an antibiotic (trade) name or abbreviation and it will search \code{x} and the \code{\link{antibiotics}} data set for any column containing a name or ATC code of that antibiotic. \strong{Longer columns names take precendence over shorter column names.}
#' @importFrom dplyr %>% select filter_all any_vars
#' @importFrom crayon blue
#' @return A column name of \code{x}, or \code{NULL} when no result is found.
#' @export
#' @inheritSection AMR Read more on our website!
#' @examples
#' df <- data.frame(amox = "S",
#' tetr = "R")
#'
#' guess_ab_col(df, "amoxicillin")
#' # [1] "amox"
#' guess_ab_col(df, "J01AA07") # ATC code of tetracycline
#' # [1] "tetr"
#'
#' guess_ab_col(df, "J01AA07", verbose = TRUE)
#' # Note: Using column `tetr` as input for "J01AA07".
#' # [1] "tetr"
#'
#' # WHONET codes
#' df <- data.frame(AMP_ND10 = "R",
#' AMC_ED20 = "S")
#' guess_ab_col(df, "ampicillin")
#' # [1] "AMP_ND10"
#' guess_ab_col(df, "J01CR02")
#' # [1] "AMC_ED20"
#' guess_ab_col(df, as.ab("augmentin"))
#' # [1] "AMC_ED20"
#'
#' # Longer names take precendence:
#' df <- data.frame(AMP_ED2 = "S",
#' AMP_ED20 = "S")
#' guess_ab_col(df, "ampicillin")
#' # [1] "AMP_ED20"
guess_ab_col <- function(x = NULL, search_string = NULL, verbose = FALSE) {
if (is.null(x) & is.null(search_string)) {
return(as.name("guess_ab_col"))
}
if (!is.data.frame(x)) {
stop("`x` must be a data.frame")
}
if (length(search_string) > 1) {
warning("argument 'search_string' has length > 1 and only the first element will be used")
search_string <- search_string[1]
}
search_string <- as.character(search_string)
if (search_string %in% colnames(x)) {
ab_result <- search_string
} else {
search_string.ab <- suppressWarnings(as.ab(search_string))
if (search_string.ab %in% colnames(x)) {
ab_result <- colnames(x)[colnames(x) == search_string.ab][1L]
} else {
# sort colnames on length - longest first
cols <- colnames(x[, x %>% colnames() %>% nchar() %>% order() %>% rev()])
df_trans <- data.frame(cols = cols,
abs = suppressWarnings(as.ab(cols)),
stringsAsFactors = FALSE)
ab_result <- df_trans[which(df_trans$abs == search_string.ab), "cols"]
ab_result <- ab_result[!is.na(ab_result)][1L]
}
}
if (length(ab_result) == 0) {
if (verbose == TRUE) {
message(paste0("No column found as input for `", search_string,
"` (", ab_name(search_string, language = "en", tolower = TRUE), ")."))
}
return(NULL)
} else {
if (verbose == TRUE) {
message(blue(paste0("NOTE: Using column `", bold(ab_result), "` as input for `", search_string,
"` (", ab_name(search_string, language = "en", tolower = TRUE), ").")))
}
return(ab_result)
}
}
#' @importFrom crayon blue bold
#' @importFrom dplyr %>% mutate arrange pull
get_column_abx <- function(x,
soft_dependencies = NULL,
hard_dependencies = NULL,
verbose = FALSE,
...) {
# determine from given data set
df_trans <- data.frame(colnames = colnames(x),
abcode = suppressWarnings(as.ab(colnames(x))))
df_trans <- df_trans[!is.na(df_trans$abcode),]
x <- as.character(df_trans$colnames)
names(x) <- df_trans$abcode
# add from self-defined dots (...):
# get_column_abx(septic_patients %>% rename(thisone = AMX), amox = "thisone")
dots <- list(...)
if (length(dots) > 0) {
newnames <- suppressWarnings(as.ab(names(dots)))
if (any(is.na(newnames))) {
warning("Invalid antibiotic reference(s): ", toString(names(dots)[is.na(newnames)]),
call. = FALSE, immediate. = TRUE)
}
# turn all NULLs to NAs
dots <- unlist(lapply(dots, function(x) if (is.null(x)) NA else x))
names(dots) <- newnames
dots <- dots[!is.na(names(dots))]
# merge, but overwrite automatically determined ones by 'dots'
x <- c(x[!x %in% dots & !names(x) %in% names(dots)], dots)
# delete NAs, this will make e.g. eucast_rules(... TMP = NULL) work to prevent TMP from being used
x <- x[!is.na(x)]
}
# sort on name
x <- x[sort(names(x))]
dupes <- x[base::duplicated(x)]
if (verbose == TRUE) {
for (i in 1:length(x)) {
if (x[i] %in% dupes) {
message(red(paste0("NOTE: Using column `", bold(x[i]), "` as input for `", names(x)[i],
"` (", ab_name(names(x)[i], language = "en", tolower = TRUE), ") [DUPLICATED USE].")))
} else {
message(blue(paste0("NOTE: Using column `", bold(x[i]), "` as input for `", names(x)[i],
"` (", ab_name(names(x)[i], language = "en", tolower = TRUE), ").")))
}
}
}
if (n_distinct(x) != length(x)) {
msg_txt <- paste("Column(s)", paste0("`", dupes, "`", collapse = " and "), "used for more than one antibiotic.")
if (verbose == FALSE) {
msg_txt <- paste(msg_txt, "Use verbose = TRUE to see which antibiotics are used by which columns.")
}
stop(msg_txt, call. = FALSE)
}
if (!is.null(hard_dependencies)) {
if (!all(hard_dependencies %in% names(x))) {
# missing a hard dependency will return NA and consequently the data will not be analysed
missing <- hard_dependencies[!hard_dependencies %in% names(x)]
generate_warning_abs_missing(missing, any = FALSE)
return(NA)
}
}
if (!is.null(soft_dependencies)) {
if (!all(soft_dependencies %in% names(x))) {
# missing a soft dependency may lower the reliability
missing <- soft_dependencies[!soft_dependencies %in% names(x)]
missing_txt <- data.frame(missing = missing,
missing_names = AMR::ab_name(missing, tolower = TRUE),
stringsAsFactors = FALSE) %>%
mutate(txt = paste0(bold(missing), " (", missing_names, ")")) %>%
arrange(missing_names) %>%
pull(txt)
message(blue('NOTE: Reliability might be improved if these antimicrobial results would be available too:',
paste(missing_txt, collapse = ", ")))
}
}
x
}
generate_warning_abs_missing <- function(missing, any = FALSE) {
missing <- paste0(missing, " (", ab_name(missing, tolower = TRUE), ")")
if (any == TRUE) {
any_txt <- c(" any of", "is")
} else {
any_txt <- c("", "are")
}
warning(paste0("Introducing NAs since", any_txt[1], " these antimicrobials ", any_txt[2], " required: ",
paste(missing, collapse = ", ")),
immediate. = TRUE,
call. = FALSE)
}