AMR/R/guess_ab_col.R

342 lines
15 KiB
R
Raw Normal View History

2019-01-03 23:56:19 +01:00
# ==================================================================== #
# TITLE #
# Antimicrobial Resistance (AMR) Data Analysis for R #
2019-01-03 23:56:19 +01:00
# #
# SOURCE #
2020-07-08 14:48:06 +02:00
# https://github.com/msberends/AMR #
2019-01-03 23:56:19 +01:00
# #
# LICENCE #
2021-12-23 18:56:28 +01:00
# (c) 2018-2022 Berends MS, Luz CF et al. #
2020-10-08 11:16:03 +02:00
# Developed at the University of Groningen, the Netherlands, in #
# collaboration with non-profit organisations Certe Medical #
# Diagnostics & Advice, and University Medical Center Groningen. #
2019-01-03 23:56:19 +01:00
# #
# 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. #
# We created this package for both routine data analysis and academic #
# research and it was publicly released in the hope that it will be #
# useful, but it comes WITHOUT ANY WARRANTY OR LIABILITY. #
2020-10-08 11:16:03 +02:00
# #
# Visit our website for the full manual and a complete tutorial about #
# how to conduct AMR data analysis: https://msberends.github.io/AMR/ #
2019-01-03 23:56:19 +01:00
# ==================================================================== #
#' Guess Antibiotic Column
2019-01-03 23:56:19 +01:00
#'
#' This tries to find a column name in a data set based on information from the [antibiotics] data set. Also supports WHONET abbreviations.
#' @inheritSection lifecycle Stable Lifecycle
#' @param x a [data.frame]
#' @param search_string a text to search `x` for, will be checked with [as.ab()] if this value is not a column in `x`
2021-05-12 18:15:03 +02:00
#' @param verbose a [logical] to indicate whether additional info should be printed
#' @param only_rsi_columns a [logical] to indicate whether only antibiotic columns must be detected that were transformed to class `<rsi>` (see [as.rsi()]) on beforehand (defaults to `FALSE`)
#' @details You can look for an antibiotic (trade) name or abbreviation and it will search `x` and the [antibiotics] data set for any column containing a name or code of that antibiotic. **Longer columns names take precedence over shorter column names.**
#' @return A column name of `x`, or `NULL` when no result is found.
2019-01-03 23:56:19 +01:00
#' @export
#' @inheritSection AMR Read more on Our Website!
2019-01-11 20:37:23 +01:00
#' @examples
#' df <- data.frame(amox = "S",
#' tetr = "R")
#'
#' guess_ab_col(df, "amoxicillin")
#' # [1] "amox"
2019-05-10 16:44:59 +02:00
#' guess_ab_col(df, "J01AA07") # ATC code of tetracycline
2019-01-11 20:37:23 +01:00
#' # [1] "tetr"
#'
#' guess_ab_col(df, "J01AA07", verbose = TRUE)
#' # NOTE: Using column 'tetr' as input for J01AA07 (tetracycline).
2019-01-11 20:37:23 +01:00
#' # [1] "tetr"
2019-01-29 00:06:50 +01:00
#'
#' # 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"
2019-05-10 16:44:59 +02:00
#' guess_ab_col(df, as.ab("augmentin"))
2019-01-29 00:06:50 +01:00
#' # [1] "AMC_ED20"
2019-05-31 14:40:15 +02:00
#'
#' # 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, only_rsi_columns = FALSE) {
meet_criteria(x, allow_class = "data.frame", allow_NULL = TRUE)
meet_criteria(search_string, allow_class = "character", has_length = 1, allow_NULL = TRUE)
meet_criteria(verbose, allow_class = "logical", has_length = 1)
meet_criteria(only_rsi_columns, allow_class = "logical", has_length = 1)
2019-05-13 10:10:16 +02:00
if (is.null(x) & is.null(search_string)) {
2019-01-11 20:37:23 +01:00
return(as.name("guess_ab_col"))
2019-05-10 16:44:59 +02:00
} else {
meet_criteria(search_string, allow_class = "character", has_length = 1, allow_NULL = FALSE)
2019-01-03 23:56:19 +01:00
}
2020-07-13 09:17:24 +02:00
all_found <- get_column_abx(x, info = verbose, only_rsi_columns = only_rsi_columns,
verbose = verbose, fn = "guess_ab_col")
search_string.ab <- suppressWarnings(as.ab(search_string))
ab_result <- unname(all_found[names(all_found) == search_string.ab])
2019-01-11 20:37:23 +01:00
if (length(ab_result) == 0) {
2019-01-03 23:56:19 +01:00
if (verbose == TRUE) {
message_("No column found as input for ", search_string,
" (", ab_name(search_string, language = NULL, tolower = TRUE), ").",
2020-10-27 15:56:51 +01:00
add_fn = font_black,
as_note = FALSE)
2019-01-03 23:56:19 +01:00
}
return(NULL)
2019-01-11 20:37:23 +01:00
} else {
if (verbose == TRUE) {
message_("Using column '", font_bold(ab_result), "' as input for ", search_string,
" (", ab_name(search_string, language = NULL, tolower = TRUE), ").")
2019-01-11 20:37:23 +01:00
}
2019-05-10 16:44:59 +02:00
return(ab_result)
2019-01-03 23:56:19 +01:00
}
}
2019-06-27 11:57:45 +02:00
get_column_abx <- function(x,
...,
2019-06-27 11:57:45 +02:00
soft_dependencies = NULL,
hard_dependencies = NULL,
verbose = FALSE,
2020-09-24 00:30:11 +02:00
info = TRUE,
only_rsi_columns = FALSE,
sort = TRUE,
reuse_previous_result = TRUE,
fn = NULL) {
2021-05-24 09:34:08 +02:00
# check if retrieved before, then get it from package environment
if (isTRUE(reuse_previous_result) && identical(unique_call_id(entire_session = FALSE,
match_fn = fn),
pkg_env$get_column_abx.call)) {
# so within the same call, within the same environment, we got here again.
# but we could've come from another function within the same call, so now only check the columns that changed
# first remove the columns that are not existing anymore
previous <- pkg_env$get_column_abx.out
current <- previous[previous %in% colnames(x)]
# then compare columns in current call with columns in original call
new_cols <- colnames(x)[!colnames(x) %in% pkg_env$get_column_abx.checked_cols]
if (length(new_cols) > 0) {
# these columns did not exist in the last call, so add them
new_cols_rsi <- get_column_abx(x[, new_cols, drop = FALSE], reuse_previous_result = FALSE, info = FALSE, sort = FALSE)
current <- c(current, new_cols_rsi)
# order according to columns in current call
current <- current[match(colnames(x)[colnames(x) %in% current], current)]
}
# update pkg environment to improve speed on next run
pkg_env$get_column_abx.out <- current
pkg_env$get_column_abx.checked_cols <- colnames(x)
# and return right values
2021-05-24 09:34:08 +02:00
return(pkg_env$get_column_abx.out)
}
meet_criteria(x, allow_class = "data.frame")
meet_criteria(soft_dependencies, allow_class = "character", allow_NULL = TRUE)
meet_criteria(hard_dependencies, allow_class = "character", allow_NULL = TRUE)
meet_criteria(verbose, allow_class = "logical", has_length = 1)
meet_criteria(info, allow_class = "logical", has_length = 1)
meet_criteria(only_rsi_columns, allow_class = "logical", has_length = 1)
meet_criteria(sort, allow_class = "logical", has_length = 1)
2020-07-13 09:17:24 +02:00
2020-09-24 00:30:11 +02:00
if (info == TRUE) {
2021-01-15 22:44:52 +01:00
message_("Auto-guessing columns suitable for analysis", appendLF = FALSE, as_note = FALSE)
2020-09-24 00:30:11 +02:00
}
2019-10-08 10:02:19 +02:00
x <- as.data.frame(x, stringsAsFactors = FALSE)
x.bak <- x
if (only_rsi_columns == TRUE) {
x <- x[, which(is.rsi(x)), drop = FALSE]
}
2020-06-09 16:18:03 +02:00
if (NROW(x) > 10000) {
# only test maximum of 10,000 values per column
2020-09-24 00:30:11 +02:00
if (info == TRUE) {
2020-10-27 15:56:51 +01:00
message_(" (using only ", font_bold("the first 10,000 rows"), ")...",
appendLF = FALSE,
as_note = FALSE)
2020-09-24 00:30:11 +02:00
}
2020-06-09 16:18:03 +02:00
x <- x[1:10000, , drop = FALSE]
2020-09-24 00:30:11 +02:00
} else if (info == TRUE) {
2020-10-27 15:56:51 +01:00
message_("...", appendLF = FALSE, as_note = FALSE)
2020-06-09 16:18:03 +02:00
}
2019-10-08 10:02:19 +02:00
# only check columns that are a valid AB code, ATC code, name, abbreviation or synonym,
# or already have the <rsi> class (as.rsi)
# and that they have no more than 50% invalid values
vectr_antibiotics <- unlist(AB_lookup$generalised_all)
2019-10-08 10:02:19 +02:00
vectr_antibiotics <- vectr_antibiotics[!is.na(vectr_antibiotics) & nchar(vectr_antibiotics) >= 3]
x_columns <- vapply(FUN.VALUE = character(1),
colnames(x),
function(col, df = x) {
if (generalise_antibiotic_name(col) %in% vectr_antibiotics ||
is.rsi(x[, col, drop = TRUE]) ||
is.rsi.eligible(x[, col, drop = TRUE], threshold = 0.5)
) {
return(col)
} else {
return(NA_character_)
}
}, USE.NAMES = FALSE)
x_columns <- x_columns[!is.na(x_columns)]
x <- x[, x_columns, drop = FALSE] # without drop = FALSE, x will become a vector when x_columns is length 1
2019-06-27 11:57:45 +02:00
df_trans <- data.frame(colnames = colnames(x),
abcode = suppressWarnings(as.ab(colnames(x), info = FALSE)),
stringsAsFactors = FALSE)
2020-09-24 00:30:11 +02:00
df_trans <- df_trans[!is.na(df_trans$abcode), , drop = FALSE]
out <- as.character(df_trans$colnames)
names(out) <- df_trans$abcode
2020-07-13 09:17:24 +02:00
2019-06-27 11:57:45 +02:00
# add from self-defined dots (...):
# such as get_column_abx(example_isolates %>% rename(thisone = AMX), amox = "thisone")
all_okay <- TRUE
2019-06-27 11:57:45 +02:00
dots <- list(...)
# remove data.frames, since this is also used running `eucast_rules(eucast_rules_df = df)`
dots <- dots[!vapply(FUN.VALUE = logical(1), dots, is.data.frame)]
2019-06-27 11:57:45 +02:00
if (length(dots) > 0) {
newnames <- suppressWarnings(as.ab(names(dots), info = FALSE))
2019-06-27 11:57:45 +02:00
if (any(is.na(newnames))) {
if (info == TRUE) {
message_(" WARNING", add_fn = list(font_yellow, font_bold), as_note = FALSE)
}
warning_("Invalid antibiotic reference(s): ", vector_and(names(dots)[is.na(newnames)], quotes = FALSE),
2020-11-10 16:35:56 +01:00
call = FALSE,
immediate = TRUE)
all_okay <- FALSE
}
unexisting_cols <- which(!vapply(FUN.VALUE = logical(1), dots, function(col) all(col %in% x_columns)))
if (length(unexisting_cols) > 0) {
if (info == TRUE) {
message_(" ERROR", add_fn = list(font_red, font_bold), as_note = FALSE)
}
stop_("Column(s) not found: ", vector_and(unlist(dots[[unexisting_cols]]), quotes = FALSE),
call = FALSE)
all_okay <- FALSE
2019-06-27 11:57:45 +02:00
}
# turn all NULLs to NAs
dots <- unlist(lapply(dots, function(dot) if (is.null(dot)) NA else dot))
2019-06-27 11:57:45 +02:00
names(dots) <- newnames
dots <- dots[!is.na(names(dots))]
# merge, but overwrite automatically determined ones by 'dots'
out <- c(out[!out %in% dots & !names(out) %in% names(dots)], dots)
2019-06-27 11:57:45 +02:00
# delete NAs, this will make e.g. eucast_rules(... TMP = NULL) work to prevent TMP from being used
out <- out[!is.na(out)]
2019-06-27 11:57:45 +02:00
}
2020-07-13 09:17:24 +02:00
if (length(out) == 0) {
if (info == TRUE & all_okay == TRUE) {
2020-10-27 15:56:51 +01:00
message_("No columns found.")
2020-09-24 00:30:11 +02:00
}
pkg_env$get_column_abx.call <- unique_call_id(entire_session = FALSE, match_fn = fn)
pkg_env$get_column_abx.checked_cols <- colnames(x.bak)
pkg_env$get_column_abx.out <- out
return(out)
2020-06-03 11:48:00 +02:00
}
2019-06-27 11:57:45 +02:00
# sort on name
if (sort == TRUE) {
out <- out[order(names(out), out)]
}
2021-12-09 10:48:25 +01:00
# only keep the first hits, no duplicates
duplicates <- c(out[duplicated(names(out))], out[duplicated(unname(out))])
if (length(duplicates) > 0) {
all_okay <- FALSE
2020-09-24 00:30:11 +02:00
}
2020-07-13 09:17:24 +02:00
if (info == TRUE) {
if (all_okay == TRUE) {
message_(" OK.", add_fn = list(font_green, font_bold), as_note = FALSE)
} else {
message_(" WARNING.", add_fn = list(font_yellow, font_bold), as_note = FALSE)
2019-06-27 11:57:45 +02:00
}
for (i in seq_len(length(out))) {
if (verbose == TRUE & !names(out[i]) %in% names(duplicates)) {
message_("Using column '", font_bold(out[i]), "' as input for ", names(out)[i],
" (", ab_name(names(out)[i], tolower = TRUE, language = NULL), ").")
}
if (names(out[i]) %in% names(duplicates)) {
already_set_as <- out[unname(out) == unname(out[i])][1L]
warning_(paste0("Column '", font_bold(out[i]), "' will not be used for ",
names(out)[i], " (", ab_name(names(out)[i], tolower = TRUE, language = NULL), ")",
", as it is already set for ",
names(already_set_as), " (", ab_name(names(already_set_as), tolower = TRUE, language = NULL), ")"),
add_fn = font_red,
immediate = verbose)
}
2019-06-27 11:57:45 +02:00
}
}
out <- out[!duplicated(names(out))]
out <- out[!duplicated(unname(out))]
if (sort == TRUE) {
out <- out[order(names(out), out)]
}
2019-06-27 11:57:45 +02:00
if (!is.null(hard_dependencies)) {
hard_dependencies <- unique(hard_dependencies)
if (!all(hard_dependencies %in% names(out))) {
2019-06-27 11:57:45 +02:00
# missing a hard dependency will return NA and consequently the data will not be analysed
missing <- hard_dependencies[!hard_dependencies %in% names(out)]
2019-06-27 11:57:45 +02:00
generate_warning_abs_missing(missing, any = FALSE)
return(NA)
}
}
if (!is.null(soft_dependencies)) {
soft_dependencies <- unique(soft_dependencies)
if (info == TRUE & !all(soft_dependencies %in% names(out))) {
2019-06-27 11:57:45 +02:00
# missing a soft dependency may lower the reliability
missing <- soft_dependencies[!soft_dependencies %in% names(out)]
missing_msg <- vector_and(paste0(ab_name(missing, tolower = TRUE, language = NULL),
" (", font_bold(missing, collapse = NULL), ")"),
quotes = FALSE)
2020-10-27 15:56:51 +01:00
message_("Reliability would be improved if these antimicrobial results would be available too: ",
missing_msg)
2019-06-27 11:57:45 +02:00
}
}
2021-05-24 09:34:08 +02:00
pkg_env$get_column_abx.call <- unique_call_id(entire_session = FALSE, match_fn = fn)
pkg_env$get_column_abx.checked_cols <- colnames(x.bak)
pkg_env$get_column_abx.out <- out
out
2019-06-27 11:57:45 +02:00
}
get_ab_from_namespace <- function(x, cols_ab) {
# cols_ab comes from get_column_abx()
x <- trimws(unique(toupper(unlist(strsplit(x, ",")))))
x_new <- character()
for (val in x) {
if (paste0("AB_", val) %in% ls(envir = asNamespace("AMR"))) {
# antibiotic group names, as defined in data-raw/_internals.R, such as `AB_CARBAPENEMS`
val <- eval(parse(text = paste0("AB_", val)), envir = asNamespace("AMR"))
} else if (val %in% AB_lookup$ab) {
# separate drugs, such as `AMX`
val <- as.ab(val)
} else {
stop_("unknown antimicrobial agent (group): ", val, call = FALSE)
}
x_new <- c(x_new, val)
}
x_new <- unique(x_new)
out <- cols_ab[match(x_new, names(cols_ab))]
out[!is.na(out)]
}
2019-06-27 11:57:45 +02:00
generate_warning_abs_missing <- function(missing, any = FALSE) {
2020-05-18 11:09:02 +02:00
missing <- paste0(missing, " (", ab_name(missing, tolower = TRUE, language = NULL), ")")
2019-06-27 11:57:45 +02:00
if (any == TRUE) {
any_txt <- c(" any of", "is")
} else {
any_txt <- c("", "are")
}
2020-11-10 16:35:56 +01:00
warning_(paste0("Introducing NAs since", any_txt[1], " these antimicrobials ", any_txt[2], " required: ",
vector_and(missing, quotes = FALSE)),
immediate = TRUE)
2019-06-27 11:57:45 +02:00
}