mirror of https://github.com/msberends/AMR.git
922 lines
40 KiB
R
Executable File
922 lines
40 KiB
R
Executable File
# ==================================================================== #
|
|
# TITLE #
|
|
# AMR: An R Package for Working with Antimicrobial Resistance Data #
|
|
# #
|
|
# SOURCE #
|
|
# https://github.com/msberends/AMR #
|
|
# #
|
|
# CITE AS #
|
|
# Berends MS, Luz CF, Friedrich AW, Sinha BNM, Albers CJ, Glasner C #
|
|
# (2022). AMR: An R Package for Working with Antimicrobial Resistance #
|
|
# Data. Journal of Statistical Software, 104(3), 1-31. #
|
|
# doi:10.18637/jss.v104.i03 #
|
|
# #
|
|
# Developed at the University of Groningen and the University Medical #
|
|
# Center Groningen in The Netherlands, in collaboration with many #
|
|
# colleagues from around the world, see our website. #
|
|
# #
|
|
# 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. #
|
|
# #
|
|
# Visit our website for the full manual and a complete tutorial about #
|
|
# how to conduct AMR data analysis: https://msberends.github.io/AMR/ #
|
|
# ==================================================================== #
|
|
|
|
#' Get Properties of a Microorganism
|
|
#'
|
|
#' Use these functions to return a specific property of a microorganism based on the latest accepted taxonomy. All input values will be evaluated internally with [as.mo()], which makes it possible to use microbial abbreviations, codes and names as input. See *Examples*.
|
|
#' @param x any [character] (vector) that can be coerced to a valid microorganism code with [as.mo()]. Can be left blank for auto-guessing the column containing microorganism codes if used in a data set, see *Examples*.
|
|
#' @param property one of the column names of the [microorganisms] data set: `r vector_or(colnames(microorganisms), sort = FALSE, quotes = TRUE)`, or must be `"shortname"`
|
|
#' @inheritParams as.mo
|
|
#' @param ... other arguments passed on to [as.mo()], such as 'minimum_matching_score', 'ignore_pattern', and 'remove_from_input'
|
|
#' @param ab any (vector of) text that can be coerced to a valid antibiotic drug code with [as.ab()]
|
|
#' @param open browse the URL using [`browseURL()`][utils::browseURL()]
|
|
#' @details All functions will, at default, **not** keep old taxonomic properties, as synonyms are automatically replaced with the current taxonomy. Take for example *Enterobacter aerogenes*, which was initially named in 1960 but renamed to *Klebsiella aerogenes* in 2017:
|
|
#' - `mo_genus("Enterobacter aerogenes")` will return `"Klebsiella"` (with a note about the renaming)
|
|
#' - `mo_genus("Enterobacter aerogenes", keep_synonyms = TRUE)` will return `"Enterobacter"` (with a once-per-session warning that the name is outdated)
|
|
#' - `mo_ref("Enterobacter aerogenes")` will return `"Tindall et al., 2017"` (with a note)
|
|
#' - `mo_ref("Enterobacter aerogenes", keep_synonyms = TRUE)` will return `"Hormaeche et al., 1960"` (with a warning)
|
|
#'
|
|
#' The short name ([mo_shortname()]) returns the first character of the genus and the full species, such as `"E. coli"`, for species and subspecies. Exceptions are abbreviations of staphylococci (such as *"CoNS"*, Coagulase-Negative Staphylococci) and beta-haemolytic streptococci (such as *"GBS"*, Group B Streptococci). Please bear in mind that e.g. *E. coli* could mean *Escherichia coli* (kingdom of Bacteria) as well as *Entamoeba coli* (kingdom of Protozoa). Returning to the full name will be done using [as.mo()] internally, giving priority to bacteria and human pathogens, i.e. `"E. coli"` will be considered *Escherichia coli*. As a result, `mo_fullname(mo_shortname("Entamoeba coli"))` returns `"Escherichia coli"`.
|
|
#'
|
|
#' Since the top-level of the taxonomy is sometimes referred to as 'kingdom' and sometimes as 'domain', the functions [mo_kingdom()] and [mo_domain()] return the exact same results.
|
|
#'
|
|
#' Determination of human pathogenicity ([mo_pathogenicity()]) is strongly based on Bartlett *et al.* (2022, \doi{10.1099/mic.0.001269}). This function returns a [factor] with the levels *Pathogenic*, *Potentially pathogenic*, *Non-pathogenic*, and *Unknown*.
|
|
#'
|
|
#' Determination of the Gram stain ([mo_gramstain()]) will be based on the taxonomic kingdom and phylum. Originally, Cavalier-Smith defined the so-called subkingdoms Negibacteria and Posibacteria (2002, [PMID 11837318](https://pubmed.ncbi.nlm.nih.gov/11837318/)), and only considered these phyla as Posibacteria: Actinobacteria, Chloroflexi, Firmicutes, and Tenericutes. These phyla were later renamed to Actinomycetota, Chloroflexota, Bacillota, and Mycoplasmatota (2021, [PMID 34694987](https://pubmed.ncbi.nlm.nih.gov/34694987/)). Bacteria in these phyla are considered Gram-positive in this `AMR` package, except for members of the class Negativicutes (within phylum Bacillota) which are Gram-negative. All other bacteria are considered Gram-negative. Species outside the kingdom of Bacteria will return a value `NA`. Functions [mo_is_gram_negative()] and [mo_is_gram_positive()] always return `TRUE` or `FALSE` (or `NA` when the input is `NA` or the MO code is `UNKNOWN`), thus always return `FALSE` for species outside the taxonomic kingdom of Bacteria.
|
|
#'
|
|
#' Determination of yeasts ([mo_is_yeast()]) will be based on the taxonomic kingdom and class. *Budding yeasts* are fungi of the phylum Ascomycota, class Saccharomycetes (also called Hemiascomycetes). *True yeasts* are aggregated into the underlying order Saccharomycetales. Thus, for all microorganisms that are member of the taxonomic class Saccharomycetes, the function will return `TRUE`. It returns `FALSE` otherwise (or `NA` when the input is `NA` or the MO code is `UNKNOWN`).
|
|
#'
|
|
#' Determination of intrinsic resistance ([mo_is_intrinsic_resistant()]) will be based on the [intrinsic_resistant] data set, which is based on `r format_eucast_version_nr(3.3)`. The [mo_is_intrinsic_resistant()] function can be vectorised over both argument `x` (input for microorganisms) and `ab` (input for antibiotics).
|
|
#'
|
|
#' The function [mo_url()] will return the direct URL to the online database entry, which also shows the scientific reference of the concerned species.
|
|
#'
|
|
#' SNOMED codes ([mo_snomed()]) are from the version of `r documentation_date(TAXONOMY_VERSION$SNOMED$accessed_date)`. See *Source* and the [microorganisms] data set for more info.
|
|
#'
|
|
#' Old taxonomic names (so-called 'synonyms') can be retrieved with [mo_synonyms()], the current taxonomic name can be retrieved with [mo_current()]. Both functions return full names.
|
|
#'
|
|
#' All output [will be translated][translate] where possible.
|
|
#' @section Matching Score for Microorganisms:
|
|
#' This function uses [as.mo()] internally, which uses an advanced algorithm to translate arbitrary user input to valid taxonomy using a so-called matching score. You can read about this public algorithm on the [MO matching score page][mo_matching_score()].
|
|
#' @inheritSection as.mo Source
|
|
#' @rdname mo_property
|
|
#' @name mo_property
|
|
#' @return
|
|
#' - An [integer] in case of [mo_year()]
|
|
#' - An [ordered factor][factor] in case of [mo_pathogenicity()]
|
|
#' - A [list] in case of [mo_taxonomy()], [mo_synonyms()] and [mo_info()]
|
|
#' - A named [character] in case of [mo_url()]
|
|
#' - A [numeric] in case of [mo_snomed()]
|
|
#' - A [character] in all other cases
|
|
#' @export
|
|
#' @seealso Data set [microorganisms]
|
|
#' @inheritSection AMR Reference Data Publicly Available
|
|
#' @examples
|
|
#' # taxonomic tree -----------------------------------------------------------
|
|
#'
|
|
#' mo_kingdom("Klebsiella pneumoniae")
|
|
#' mo_phylum("Klebsiella pneumoniae")
|
|
#' mo_class("Klebsiella pneumoniae")
|
|
#' mo_order("Klebsiella pneumoniae")
|
|
#' mo_family("Klebsiella pneumoniae")
|
|
#' mo_genus("Klebsiella pneumoniae")
|
|
#' mo_species("Klebsiella pneumoniae")
|
|
#' mo_subspecies("Klebsiella pneumoniae")
|
|
#'
|
|
#'
|
|
#' # full names and short names -----------------------------------------------
|
|
#'
|
|
#' mo_name("Klebsiella pneumoniae")
|
|
#' mo_fullname("Klebsiella pneumoniae")
|
|
#' mo_shortname("Klebsiella pneumoniae")
|
|
#'
|
|
#'
|
|
#' # other properties ---------------------------------------------------------
|
|
#'
|
|
#' mo_pathogenicity("Klebsiella pneumoniae")
|
|
#' mo_gramstain("Klebsiella pneumoniae")
|
|
#' mo_snomed("Klebsiella pneumoniae")
|
|
#' mo_type("Klebsiella pneumoniae")
|
|
#' mo_rank("Klebsiella pneumoniae")
|
|
#' mo_url("Klebsiella pneumoniae")
|
|
#' mo_is_yeast(c("Candida", "Trichophyton", "Klebsiella"))
|
|
#'
|
|
#'
|
|
#' # scientific reference -----------------------------------------------------
|
|
#'
|
|
#' mo_ref("Klebsiella pneumoniae")
|
|
#' mo_authors("Klebsiella pneumoniae")
|
|
#' mo_year("Klebsiella pneumoniae")
|
|
#' mo_lpsn("Klebsiella pneumoniae")
|
|
#' mo_gbif("Klebsiella pneumoniae")
|
|
#' mo_synonyms("Klebsiella pneumoniae")
|
|
#'
|
|
#'
|
|
#' # abbreviations known in the field -----------------------------------------
|
|
#'
|
|
#' mo_genus("MRSA")
|
|
#' mo_species("MRSA")
|
|
#' mo_shortname("VISA")
|
|
#' mo_gramstain("VISA")
|
|
#'
|
|
#' mo_genus("EHEC")
|
|
#' mo_species("EHEC")
|
|
#'
|
|
#'
|
|
#' # known subspecies ---------------------------------------------------------
|
|
#'
|
|
#' mo_fullname("K. pneu rh")
|
|
#' mo_shortname("K. pneu rh")
|
|
#'
|
|
#'
|
|
#' \donttest{
|
|
#' # Becker classification, see ?as.mo ----------------------------------------
|
|
#'
|
|
#' mo_fullname("Staph epidermidis")
|
|
#' mo_fullname("Staph epidermidis", Becker = TRUE)
|
|
#' mo_shortname("Staph epidermidis")
|
|
#' mo_shortname("Staph epidermidis", Becker = TRUE)
|
|
#'
|
|
#'
|
|
#' # Lancefield classification, see ?as.mo ------------------------------------
|
|
#'
|
|
#' mo_fullname("Strep agalactiae")
|
|
#' mo_fullname("Strep agalactiae", Lancefield = TRUE)
|
|
#' mo_shortname("Strep agalactiae")
|
|
#' mo_shortname("Strep agalactiae", Lancefield = TRUE)
|
|
#'
|
|
#'
|
|
#' # language support --------------------------------------------------------
|
|
#'
|
|
#' mo_gramstain("Klebsiella pneumoniae", language = "de") # German
|
|
#' mo_gramstain("Klebsiella pneumoniae", language = "nl") # Dutch
|
|
#' mo_gramstain("Klebsiella pneumoniae", language = "es") # Spanish
|
|
#' mo_gramstain("Klebsiella pneumoniae", language = "el") # Greek
|
|
#' mo_gramstain("Klebsiella pneumoniae", language = "uk") # Ukrainian
|
|
#'
|
|
#' # mo_type is equal to mo_kingdom, but mo_kingdom will remain untranslated
|
|
#' mo_kingdom("Klebsiella pneumoniae")
|
|
#' mo_type("Klebsiella pneumoniae")
|
|
#' mo_kingdom("Klebsiella pneumoniae", language = "zh") # Chinese, no effect
|
|
#' mo_type("Klebsiella pneumoniae", language = "zh") # Chinese, translated
|
|
#'
|
|
#' mo_fullname("S. pyogenes", Lancefield = TRUE, language = "de")
|
|
#' mo_fullname("S. pyogenes", Lancefield = TRUE, language = "uk")
|
|
#'
|
|
#'
|
|
#' # other --------------------------------------------------------------------
|
|
#'
|
|
#' # gram stains and intrinsic resistance can be used as a filter in dplyr verbs
|
|
#' if (require("dplyr")) {
|
|
#' example_isolates %>%
|
|
#' filter(mo_is_gram_positive()) %>%
|
|
#' count(mo_genus(), sort = TRUE)
|
|
#' }
|
|
#' if (require("dplyr")) {
|
|
#' example_isolates %>%
|
|
#' filter(mo_is_intrinsic_resistant(ab = "vanco")) %>%
|
|
#' count(mo_genus(), sort = TRUE)
|
|
#' }
|
|
#'
|
|
#' # get a list with the complete taxonomy (from kingdom to subspecies)
|
|
#' mo_taxonomy("Klebsiella pneumoniae")
|
|
#'
|
|
#' # get a list with the taxonomy, the authors, Gram-stain,
|
|
#' # SNOMED codes, and URL to the online database
|
|
#' mo_info("Klebsiella pneumoniae")
|
|
#' }
|
|
mo_name <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_name")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "fullname", language = language, keep_synonyms = keep_synonyms, ...),
|
|
language = language,
|
|
only_unknown = FALSE,
|
|
only_affect_mo_names = TRUE
|
|
)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_fullname <- mo_name
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_shortname <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_shortname")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
replace_empty <- function(x) {
|
|
x[x == ""] <- "spp."
|
|
x
|
|
}
|
|
|
|
# get first char of genus and complete species in English
|
|
genera <- mo_genus(x.mo, language = NULL, keep_synonyms = keep_synonyms)
|
|
shortnames <- paste0(substr(genera, 1, 1), ". ", replace_empty(mo_species(x.mo, language = NULL, keep_synonyms = keep_synonyms)))
|
|
|
|
# exceptions for where no species is known
|
|
shortnames[shortnames %like% ".[.] spp[.]"] <- genera[shortnames %like% ".[.] spp[.]"]
|
|
# exceptions for staphylococci
|
|
shortnames[shortnames == "S. coagulase-negative"] <- "CoNS"
|
|
shortnames[shortnames == "S. coagulase-positive"] <- "CoPS"
|
|
# exceptions for streptococci: Group A Streptococcus -> GAS
|
|
shortnames[shortnames %like_case% "S. Group [ABCDFGHK]"] <- paste0("G", gsub("S. Group ([ABCDFGHK])", "\\1", shortnames[shortnames %like_case% "S. Group [ABCDFGHK]"], perl = TRUE), "S")
|
|
# unknown species etc.
|
|
shortnames[shortnames %like% "unknown"] <- paste0("(", trimws2(gsub("[^a-zA-Z -]", "", shortnames[shortnames %like% "unknown"], perl = TRUE)), ")")
|
|
|
|
shortnames[mo_rank(x.mo) %in% c("kingdom", "phylum", "class", "order", "family")] <- mo_name(x.mo, language = NULL, keep_synonyms = keep_synonyms)
|
|
|
|
shortnames[is.na(x.mo)] <- NA_character_
|
|
load_mo_uncertainties(metadata)
|
|
translate_into_language(shortnames, language = language, only_unknown = FALSE, only_affect_mo_names = TRUE)
|
|
}
|
|
|
|
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_subspecies <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_subspecies")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "subspecies", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_species <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_species")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "species", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_genus <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_genus")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "genus", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_family <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_family")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "family", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_order <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_order")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "order", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_class <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_class")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "class", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_phylum <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_phylum")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "phylum", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_kingdom <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_kingdom")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "kingdom", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_domain <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_domain")
|
|
}
|
|
mo_kingdom(x = x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_type <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_type")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
out <- mo_kingdom(x.mo, language = NULL, keep_synonyms = keep_synonyms)
|
|
out[which(mo_is_yeast(x.mo, keep_synonyms = keep_synonyms))] <- "Yeasts"
|
|
translate_into_language(out, language = language, only_unknown = FALSE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_status <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_status")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = "status", language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_pathogenicity <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_pathogenicity")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
prev <- AMR_env$MO_lookup$prevalence[match(x.mo, AMR_env$MO_lookup$mo)]
|
|
kngd <- AMR_env$MO_lookup$kingdom[match(x.mo, AMR_env$MO_lookup$mo)]
|
|
rank <- AMR_env$MO_lookup$rank[match(x.mo, AMR_env$MO_lookup$mo)]
|
|
|
|
out <- factor(ifelse(prev == 1 & kngd == "Bacteria" & rank != "genus",
|
|
"Pathogenic",
|
|
ifelse(prev < 2 & kngd == "Fungi",
|
|
"Potentially pathogenic",
|
|
ifelse(prev == 2 & kngd == "Bacteria",
|
|
"Non-pathogenic",
|
|
ifelse(kngd == "Bacteria",
|
|
"Potentially pathogenic",
|
|
"Unknown")))),
|
|
levels = c("Pathogenic", "Potentially pathogenic", "Non-pathogenic", "Unknown"),
|
|
ordered = TRUE)
|
|
|
|
load_mo_uncertainties(metadata)
|
|
out
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_gramstain <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_gramstain")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
x <- rep(NA_character_, length(x))
|
|
# make all bacteria Gram negative
|
|
x[mo_kingdom(x.mo, language = NULL, keep_synonyms = keep_synonyms) == "Bacteria"] <- "Gram-negative"
|
|
# overwrite these 4 phyla with Gram-positives
|
|
# Source: https://itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=956097 (Cavalier-Smith, 2002)
|
|
x[(mo_phylum(x.mo, language = NULL, keep_synonyms = keep_synonyms) %in% c(
|
|
# no longer in use, does not hurt to keep here:
|
|
"Actinobacteria",
|
|
"Chloroflexi",
|
|
"Firmicutes",
|
|
"Tenericutes",
|
|
"Actinomycetota", # since 2021, old name was Actinobacteria
|
|
"Chloroflexota", # since 2021, old name was Chloroflexi
|
|
"Bacillota", # since 2021, old name was Firmicutes
|
|
"Mycoplasmatota" # since 2021, old name was Tenericutes
|
|
) &
|
|
# but class Negativicutes (of phylum Bacillota) are Gram-negative!
|
|
mo_class(x.mo, language = NULL, keep_synonyms = keep_synonyms) != "Negativicutes")
|
|
# and of course our own ID for Gram-positives
|
|
| x.mo == "B_GRAMP"] <- "Gram-positive"
|
|
|
|
load_mo_uncertainties(metadata)
|
|
translate_into_language(x, language = language, only_unknown = FALSE)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_is_gram_negative <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_is_gram_negative")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
grams <- mo_gramstain(x.mo, language = NULL, keep_synonyms = keep_synonyms)
|
|
load_mo_uncertainties(metadata)
|
|
out <- grams == "Gram-negative" & !is.na(grams)
|
|
out[x.mo %in% c(NA_character_, "UNKNOWN")] <- NA
|
|
out
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_is_gram_positive <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_is_gram_positive")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
grams <- mo_gramstain(x.mo, language = NULL, keep_synonyms = keep_synonyms)
|
|
load_mo_uncertainties(metadata)
|
|
out <- grams == "Gram-positive" & !is.na(grams)
|
|
out[x.mo %in% c(NA_character_, "UNKNOWN")] <- NA
|
|
out
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_is_yeast <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_is_yeast")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
x.kingdom <- mo_kingdom(x.mo, language = NULL, keep_synonyms = keep_synonyms)
|
|
x.class <- mo_class(x.mo, language = NULL, keep_synonyms = keep_synonyms)
|
|
|
|
load_mo_uncertainties(metadata)
|
|
|
|
out <- rep(FALSE, length(x))
|
|
out[x.kingdom == "Fungi" & x.class == "Saccharomycetes"] <- TRUE
|
|
out[x.mo %in% c(NA_character_, "UNKNOWN")] <- NA
|
|
out
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_is_intrinsic_resistant <- function(x, ab, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_is_intrinsic_resistant")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(ab, allow_NA = FALSE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
ab <- as.ab(ab, language = NULL, flag_multiple_results = FALSE, info = FALSE)
|
|
|
|
if (length(x) == 1 & length(ab) > 1) {
|
|
x <- rep(x, length(ab))
|
|
} else if (length(ab) == 1 & length(x) > 1) {
|
|
ab <- rep(ab, length(x))
|
|
}
|
|
if (length(x) != length(ab)) {
|
|
stop_("length of `x` and `ab` must be equal, or one of them must be of length 1.")
|
|
}
|
|
|
|
# show used version number once per session (AMR_env will reload every session)
|
|
if (message_not_thrown_before("mo_is_intrinsic_resistant", "version.mo", entire_session = TRUE)) {
|
|
message_(
|
|
"Determining intrinsic resistance based on ",
|
|
format_eucast_version_nr(3.3, markdown = FALSE), ". ",
|
|
font_red("This note will be shown once per session.")
|
|
)
|
|
}
|
|
|
|
# runs against internal vector: intrinsic_resistant (see zzz.R)
|
|
add_intrinsic_resistance_to_AMR_env()
|
|
paste(x, ab) %in% AMR_env$intrinsic_resistant
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_snomed <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_snomed")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
mo_validate(x = x, property = "snomed", language = language, keep_synonyms = keep_synonyms, ...)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_ref <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_ref")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
mo_validate(x = x, property = "ref", language = language, keep_synonyms = keep_synonyms, ...)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_authors <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_authors")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x <- mo_validate(x = x, property = "ref", language = language, keep_synonyms = keep_synonyms, ...)
|
|
# remove last 4 digits and presumably the comma and space that preceed them
|
|
x[!is.na(x)] <- gsub(",? ?[0-9]{4}", "", x[!is.na(x)], perl = TRUE)
|
|
suppressWarnings(x)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_year <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_year")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x <- mo_validate(x = x, property = "ref", language = language, keep_synonyms = keep_synonyms, ...)
|
|
# get last 4 digits
|
|
x[!is.na(x)] <- gsub(".*([0-9]{4})$", "\\1", x[!is.na(x)], perl = TRUE)
|
|
suppressWarnings(as.integer(x))
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_lpsn <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_lpsn")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
mo_validate(x = x, property = "lpsn", language = language, keep_synonyms = keep_synonyms, ...)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_gbif <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_gbif")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
mo_validate(x = x, property = "gbif", language = language, keep_synonyms = keep_synonyms, ...)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_rank <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_rank")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
mo_validate(x = x, property = "rank", language = language, keep_synonyms = keep_synonyms, ...)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_taxonomy <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_taxonomy")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
out <- list(
|
|
kingdom = mo_kingdom(x, language = language, keep_synonyms = keep_synonyms),
|
|
phylum = mo_phylum(x, language = language, keep_synonyms = keep_synonyms),
|
|
class = mo_class(x, language = language, keep_synonyms = keep_synonyms),
|
|
order = mo_order(x, language = language, keep_synonyms = keep_synonyms),
|
|
family = mo_family(x, language = language, keep_synonyms = keep_synonyms),
|
|
genus = mo_genus(x, language = language, keep_synonyms = keep_synonyms),
|
|
species = mo_species(x, language = language, keep_synonyms = keep_synonyms),
|
|
subspecies = mo_subspecies(x, language = language, keep_synonyms = keep_synonyms)
|
|
)
|
|
|
|
load_mo_uncertainties(metadata)
|
|
out
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_synonyms <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_synonyms")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
syns <- lapply(x.mo, function(y) {
|
|
gbif <- AMR_env$MO_lookup$gbif[match(y, AMR_env$MO_lookup$mo)]
|
|
lpsn <- AMR_env$MO_lookup$lpsn[match(y, AMR_env$MO_lookup$mo)]
|
|
out <- AMR_env$MO_lookup[which(AMR_env$MO_lookup$lpsn_renamed_to == lpsn | AMR_env$MO_lookup$gbif_renamed_to == gbif), "fullname", drop = TRUE]
|
|
if (length(out) == 0) {
|
|
NULL
|
|
} else {
|
|
out
|
|
}
|
|
})
|
|
|
|
if (length(syns) > 1) {
|
|
names(syns) <- mo_name(x, language = language)
|
|
result <- syns
|
|
} else {
|
|
result <- unlist(syns)
|
|
}
|
|
|
|
load_mo_uncertainties(metadata)
|
|
result
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_current <- function(x, language = get_AMR_locale(), ...) {
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
x.mo <- suppressWarnings(as.mo(x, keep_synonyms = TRUE, ...))
|
|
out <- synonym_mo_to_accepted_mo(x.mo, fill_in_accepted = TRUE)
|
|
mo_name(out, language = language)
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_info <- function(x, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_info")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
info <- lapply(x, function(y) {
|
|
c(
|
|
mo_taxonomy(y, language = language, keep_synonyms = keep_synonyms),
|
|
list(
|
|
status = mo_status(y, language = language, keep_synonyms = keep_synonyms),
|
|
synonyms = mo_synonyms(y, keep_synonyms = keep_synonyms),
|
|
gramstain = mo_gramstain(y, language = language, keep_synonyms = keep_synonyms),
|
|
url = unname(mo_url(y, open = FALSE, keep_synonyms = keep_synonyms)),
|
|
ref = mo_ref(y, keep_synonyms = keep_synonyms),
|
|
snomed = unlist(mo_snomed(y, keep_synonyms = keep_synonyms))
|
|
)
|
|
)
|
|
})
|
|
if (length(info) > 1) {
|
|
names(info) <- mo_name(x)
|
|
result <- info
|
|
} else {
|
|
result <- info[[1L]]
|
|
}
|
|
|
|
load_mo_uncertainties(metadata)
|
|
result
|
|
}
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_url <- function(x, open = FALSE, language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_url")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(open, allow_class = "logical", has_length = 1)
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
x.mo <- as.mo(x = x, language = language, keep_synonyms = keep_synonyms, ... = ...)
|
|
metadata <- get_mo_uncertainties()
|
|
|
|
x.rank <- AMR_env$MO_lookup$rank[match(x.mo, AMR_env$MO_lookup$mo)]
|
|
x.name <- AMR_env$MO_lookup$fullname[match(x.mo, AMR_env$MO_lookup$mo)]
|
|
x.lpsn <- AMR_env$MO_lookup$lpsn[match(x.mo, AMR_env$MO_lookup$mo)]
|
|
x.gbif <- AMR_env$MO_lookup$gbif[match(x.mo, AMR_env$MO_lookup$mo)]
|
|
|
|
u <- character(length(x))
|
|
u[!is.na(x.gbif)] <- paste0(TAXONOMY_VERSION$GBIF$url, "/species/", x.gbif[!is.na(x.gbif)])
|
|
# overwrite with LPSN:
|
|
u[!is.na(x.lpsn)] <- paste0(TAXONOMY_VERSION$LPSN$url, "/", x.rank[!is.na(x.lpsn)], "/", gsub(" ", "-", tolower(x.name[!is.na(x.lpsn)]), fixed = TRUE))
|
|
|
|
names(u) <- x.name
|
|
|
|
if (isTRUE(open)) {
|
|
if (length(u) > 1) {
|
|
warning_("in `mo_url()`: only the first URL will be opened, as `browseURL()` only suports one string.")
|
|
}
|
|
utils::browseURL(u[1L])
|
|
}
|
|
|
|
load_mo_uncertainties(metadata)
|
|
u
|
|
}
|
|
|
|
|
|
#' @rdname mo_property
|
|
#' @export
|
|
mo_property <- function(x, property = "fullname", language = get_AMR_locale(), keep_synonyms = getOption("AMR_keep_synonyms", FALSE), ...) {
|
|
if (missing(x)) {
|
|
# this tries to find the data and an 'mo' column
|
|
x <- find_mo_col(fn = "mo_property")
|
|
}
|
|
meet_criteria(x, allow_NA = TRUE)
|
|
meet_criteria(property, allow_class = "character", has_length = 1, is_in = colnames(AMR_env$MO_lookup))
|
|
language <- validate_language(language)
|
|
meet_criteria(keep_synonyms, allow_class = "logical", has_length = 1)
|
|
|
|
translate_into_language(mo_validate(x = x, property = property, language = language, keep_synonyms = keep_synonyms, ...), language = language, only_unknown = TRUE)
|
|
}
|
|
|
|
mo_validate <- function(x, property, language, keep_synonyms = keep_synonyms, ...) {
|
|
|
|
# try to catch an error when inputting an invalid argument
|
|
# so the 'call.' can be set to FALSE
|
|
tryCatch(x[1L] %in% unlist(AMR_env$MO_lookup[1, property, drop = TRUE]),
|
|
error = function(e) stop(e$message, call. = FALSE)
|
|
)
|
|
|
|
dots <- list(...)
|
|
Becker <- dots$Becker
|
|
if (is.null(Becker) || property %in% c("kingdom", "phylum", "class", "order", "family", "genus")) {
|
|
Becker <- FALSE
|
|
}
|
|
Lancefield <- dots$Lancefield
|
|
if (is.null(Lancefield) || property %in% c("kingdom", "phylum", "class", "order", "family", "genus")) {
|
|
Lancefield <- FALSE
|
|
}
|
|
has_Becker_or_Lancefield <- Becker %in% c(TRUE, "all") || Lancefield %in% c(TRUE, "all")
|
|
|
|
# get microorganisms data set, but remove synonyms if keep_synonyms is FALSE
|
|
mo_data_check <- AMR_env$MO_lookup[which(AMR_env$MO_lookup$status %in% if (isTRUE(keep_synonyms)) c("synonym", "accepted") else "accepted"), , drop = FALSE]
|
|
|
|
if (all(x %in% c(mo_data_check$mo, NA)) && !has_Becker_or_Lancefield) {
|
|
# do nothing, just don't run the other if-else's
|
|
} else if (all(x %in% c(unlist(mo_data_check[[property]]), NA)) && !has_Becker_or_Lancefield) {
|
|
# no need to do anything, just return it
|
|
return(x)
|
|
} else {
|
|
# we need to get MO codes now
|
|
x <- replace_old_mo_codes(x, property = property)
|
|
x <- as.mo(x, language = language, keep_synonyms = keep_synonyms, ...)
|
|
}
|
|
|
|
# get property reeaaally fast using match()
|
|
x <- AMR_env$MO_lookup[[property]][match(x, AMR_env$MO_lookup$mo)]
|
|
|
|
if (property == "mo") {
|
|
return(set_clean_class(x, new_class = c("mo", "character")))
|
|
} else if (property == "snomed") {
|
|
return(sort(as.character(eval(parse(text = x)))))
|
|
} else {
|
|
# everything else as character
|
|
return(as.character(x))
|
|
}
|
|
}
|
|
|
|
find_mo_col <- function(fn) {
|
|
# this function tries to find an mo column in the data the function was called in,
|
|
# which is useful when functions are used within dplyr verbs
|
|
df <- get_current_data(arg_name = "x", call = -3) # will return an error if not found
|
|
mo <- NULL
|
|
try(
|
|
{
|
|
mo <- suppressMessages(search_type_in_df(df, "mo"))
|
|
},
|
|
silent = TRUE
|
|
)
|
|
if (!is.null(df) && !is.null(mo) && is.data.frame(df)) {
|
|
if (message_not_thrown_before(fn = fn)) {
|
|
message_("Using column '", font_bold(mo), "' as input for `", fn, "()`")
|
|
}
|
|
return(df[, mo, drop = TRUE])
|
|
} else {
|
|
stop_("argument `x` is missing and no column with info about microorganisms could be found.", call = -2)
|
|
}
|
|
}
|