# ==================================================================== # # 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.gitab.io/AMR. # # ==================================================================== # #' Property of a microorganism #' #' Use these functions to return a specific property of a microorganism from the \code{\link{microorganisms}} data set. All input values will be evaluated internally with \code{\link{as.mo}}. #' @param x any (vector of) text that can be coerced to a valid microorganism code with \code{\link{as.mo}} #' @param property one of the column names of one of the \code{\link{microorganisms}} data set or \code{"shortname"} #' @param language language of the returned text, defaults to system language (see \code{\link{get_locale}}) and can also be set with \code{\link{getOption}("AMR_locale")}. Use \code{language = NULL} or \code{language = ""} to prevent translation. #' @param ... other parameters passed on to \code{\link{as.mo}} #' @details All functions will return the most recently known taxonomic property according to the Catalogue of Life, except for \code{mo_ref}, \code{mo_authors} and \code{mo_year}. This leads to the following results: #' \itemize{ #' \item{\code{mo_fullname("Chlamydia psittaci")} will return \code{"Chlamydophila psittaci"} (with a warning about the renaming)} #' \item{\code{mo_ref("Chlamydia psittaci")} will return \code{"Page, 1968"} (with a warning about the renaming)} #' \item{\code{mo_ref("Chlamydophila psittaci")} will return \code{"Everett et al., 1999"} (without a warning)} #' } #' #' The Gram stain - \code{mo_gramstain()} - will be determined on the taxonomic kingdom and phylum. According to Cavalier-Smith (2002) who defined subkingdoms Negibacteria and Posibacteria, only these phyla are Posibacteria: Actinobacteria, Chloroflexi, Firmicutes and Tenericutes (ref: \url{https://itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=956097}). These bacteria are considered Gram positive - all other bacteria are considered Gram negative. Species outside the kingdom of Bacteria will return a value \code{NA}. #' #' The function \code{mo_url()} will return the direct URL to the species in the Catalogue of Life. #' @inheritSection get_locale Supported languages #' @inheritSection catalogue_of_life Catalogue of Life #' @inheritSection as.mo Source #' @rdname mo_property #' @name mo_property #' @return \itemize{ #' \item{An \code{integer} in case of \code{mo_year}} #' \item{A \code{list} in case of \code{mo_taxonomy}} #' \item{A \code{character} in all other cases} #' } #' @export #' @seealso \code{\link{microorganisms}} #' @inheritSection AMR Read more on our website! #' @examples #' # All properties of Escherichia coli #' ## taxonomic properties #' mo_kingdom("E. coli") # "Bacteria" #' mo_phylum("E. coli") # "Proteobacteria" #' mo_class("E. coli") # "Gammaproteobacteria" #' mo_order("E. coli") # "Enterobacteriales" #' mo_family("E. coli") # "Enterobacteriaceae" #' mo_genus("E. coli") # "Escherichia" #' mo_species("E. coli") # "coli" #' mo_subspecies("E. coli") # "" #' #' ## colloquial properties #' mo_fullname("E. coli") # "Escherichia coli" #' mo_shortname("E. coli") # "E. coli" #' #' ## other properties #' mo_gramstain("E. coli") # "Gram negative" #' mo_type("E. coli") # "Bacteria" (equal to kingdom) #' #' ## scientific reference #' mo_ref("E. coli") # "Castellani and Chalmers, 1919" #' mo_authors("E. coli") # "Castellani and Chalmers" #' mo_year("E. coli") # 1919 #' #' #' # Abbreviations known in the field #' mo_genus("MRSA") # "Staphylococcus" #' mo_species("MRSA") # "aureus" #' mo_shortname("MRSA") # "S. aureus" #' mo_gramstain("MRSA") # "Gram positive" #' #' mo_genus("VISA") # "Staphylococcus" #' mo_species("VISA") # "aureus" #' #' #' # Known subspecies #' mo_genus("doylei") # "Campylobacter" #' mo_species("doylei") # "jejuni" #' mo_fullname("doylei") # "Campylobacter jejuni doylei" #' #' mo_fullname("K. pneu rh") # "Klebsiella pneumoniae rhinoscleromatis" #' mo_shortname("K. pneu rh") # "K. pneumoniae" #' #' #' # Becker classification, see ?as.mo #' mo_fullname("S. epi") # "Staphylococcus epidermidis" #' mo_fullname("S. epi", Becker = TRUE) # "Coagulase Negative Staphylococcus (CoNS)" #' mo_shortname("S. epi") # "S. epidermidis" #' mo_shortname("S. epi", Becker = TRUE) # "CoNS" #' #' # Lancefield classification, see ?as.mo #' mo_fullname("S. pyo") # "Streptococcus pyogenes" #' mo_fullname("S. pyo", Lancefield = TRUE) # "Streptococcus group A" #' mo_shortname("S. pyo") # "S. pyogenes" #' mo_shortname("S. pyo", Lancefield = TRUE) # "GAS" ('Group A streptococci') #' #' #' # Language support for German, Dutch, Spanish, Portuguese, Italian and French #' mo_gramstain("E. coli", language = "de") # "Gramnegativ" #' mo_gramstain("E. coli", language = "nl") # "Gram-negatief" #' mo_gramstain("E. coli", language = "es") # "Gram negativo" #' #' # mo_type is equal to mo_kingdom, but mo_kingdom will remain official #' mo_kingdom("E. coli") # "Bacteria" on a German system #' mo_type("E. coli") # "Bakterien" on a German system #' mo_type("E. coli") # "Bacteria" on an English system #' #' mo_fullname("S. pyogenes", #' Lancefield = TRUE, #' language = "de") # "Streptococcus Gruppe A" #' mo_fullname("S. pyogenes", #' Lancefield = TRUE, #' language = "nl") # "Streptococcus groep A" #' #' #' # Get a list with the complete taxonomy (kingdom to subspecies) #' mo_taxonomy("E. coli") mo_fullname <- function(x, language = get_locale(), ...) { x <- mo_validate(x = x, property = "fullname", ...) mo_translate(x, language = language) } #' @rdname mo_property #' @importFrom dplyr %>% left_join mutate pull #' @export mo_shortname <- function(x, language = get_locale(), ...) { dots <- list(...) Becker <- dots$Becker if (is.null(Becker)) { Becker <- FALSE } Lancefield <- dots$Lancefield if (is.null(Lancefield)) { Lancefield <- FALSE } if (Becker %in% c(TRUE, "all") | Lancefield == TRUE) { res1 <- AMR::as.mo(x, Becker = FALSE, Lancefield = FALSE, reference_df = dots$reference_df) res2 <- suppressWarnings(AMR::as.mo(res1, ...)) res2_fullname <- mo_fullname(res2, language = language) res2_fullname[res2_fullname %like% " \\(CoNS\\)"] <- "CoNS" res2_fullname[res2_fullname %like% " \\(CoPS\\)"] <- "CoPS" res2_fullname[res2_fullname %like% " \\(KNS\\)"] <- "KNS" res2_fullname[res2_fullname %like% " \\(KPS\\)"] <- "KPS" res2_fullname[res2_fullname %like% " \\(CNS\\)"] <- "CNS" res2_fullname[res2_fullname %like% " \\(CPS\\)"] <- "CPS" res2_fullname <- gsub("Streptococcus (group|Gruppe|gruppe|groep|grupo|gruppo|groupe) (.)", "G\\2S", res2_fullname) # turn "Streptococcus group A" and "Streptococcus grupo A" to "GAS" res2_fullname_vector <- res2_fullname[res2_fullname == mo_fullname(res1)] res2_fullname[res2_fullname == mo_fullname(res1)] <- paste0(substr(mo_genus(res2_fullname_vector), 1, 1), ". ", suppressWarnings(mo_species(res2_fullname_vector))) if (sum(res1 == res2, na.rm = TRUE) > 0) { res1[res1 == res2] <- paste0(substr(mo_genus(res1[res1 == res2]), 1, 1), ". ", suppressWarnings(mo_species(res1[res1 == res2]))) } res1[res1 != res2] <- res2_fullname result <- as.character(res1) } else { x <- AMR::as.mo(x, ...) suppressWarnings( result <- data.frame(mo = x) %>% left_join(AMR::microorganisms, by = "mo") %>% mutate(shortname = ifelse(!is.na(genus) & !is.na(species), paste0(substr(genus, 1, 1), ". ", species), NA_character_)) %>% pull(shortname) ) } mo_translate(result, language = language) } #' @rdname mo_property #' @export mo_subspecies <- function(x, language = get_locale(), ...) { mo_translate(mo_validate(x = x, property = "subspecies", ...), language = language) } #' @rdname mo_property #' @export mo_species <- function(x, language = get_locale(), ...) { mo_translate(mo_validate(x = x, property = "species", ...), language = language) } #' @rdname mo_property #' @export mo_genus <- function(x, language = get_locale(), ...) { mo_translate(mo_validate(x = x, property = "genus", ...), language = language) } #' @rdname mo_property #' @export mo_family <- function(x, ...) { mo_validate(x = x, property = "family", ...) } #' @rdname mo_property #' @export mo_order <- function(x, ...) { mo_validate(x = x, property = "order", ...) } #' @rdname mo_property #' @export mo_class <- function(x, ...) { mo_validate(x = x, property = "class", ...) } #' @rdname mo_property #' @export mo_phylum <- function(x, ...) { mo_validate(x = x, property = "phylum", ...) } #' @rdname mo_property #' @export mo_kingdom <- function(x, ...) { mo_validate(x = x, property = "kingdom", ...) } #' @rdname mo_property #' @export mo_type <- function(x, language = get_locale(), ...) { mo_translate(mo_validate(x = x, property = "kingdom", ...), language = language) } #' @rdname mo_property #' @export mo_gramstain <- function(x, language = get_locale(), ...) { x.bak <- x x <- mo_phylum(x, ...) x[x %in% c("Actinobacteria", "Chloroflexi", "Firmicutes", "Tenericutes")] <- "Gram positive" x[x != "Gram positive"] <- "Gram negative" x[mo_kingdom(x.bak) != "Bacteria"] <- NA_character_ mo_translate(x, language = language) } #' @rdname mo_property #' @export mo_ref <- function(x, ...) { mo_validate(x = x, property = "ref", ...) } #' @rdname mo_property #' @export mo_authors <- function(x, ...) { x <- mo_validate(x = x, property = "ref", ...) # 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)]) x } #' @rdname mo_property #' @export mo_year <- function(x, ...) { x <- mo_validate(x = x, property = "ref", ...) # get last 4 digits x[!is.na(x)] <- gsub(".*([0-9]{4})$", "\\1", x[!is.na(x)]) as.integer(x) } #' @rdname mo_property #' @export mo_taxonomy <- function(x, ...) { x <- AMR::as.mo(x, ...) base::list(kingdom = mo_kingdom(x), phylum = mo_phylum(x), class = mo_class(x), order = mo_order(x), family = mo_family(x), genus = mo_genus(x), species = mo_species(x), subspecies = mo_subspecies(x)) } #' @rdname mo_property #' @export mo_url <- function(x, ...) { u <- mo_validate(x = x, property = "species_id", ...) u[u != ""] <- paste0(catalogue_of_life$url, "/details/species/id/", u) u } #' @rdname mo_property #' @importFrom data.table data.table as.data.table setkey #' @export mo_property <- function(x, property = 'fullname', language = get_locale(), ...) { if (length(property) != 1L) { stop("'property' must be of length 1.") } if (!property %in% colnames(AMR::microorganisms)) { stop("invalid property: '", property, "' - use a column name of the `microorganisms` data set") } mo_translate(mo_validate(x = x, property = property, ...), language = language) } #' @importFrom dplyr %>% case_when mo_translate <- function(x, language) { if (is.null(language)) { return(x) } if (language %in% c("en", "")) { return(x) } supported <- c("en", "de", "nl", "es", "pt", "it", "fr") if (!language %in% supported) { stop("Unsupported language: '", language, "' - use one of: ", paste0("'", sort(supported), "'", collapse = ", "), call. = FALSE) } x_tobetranslated <- grepl(x = x, pattern = "(Coagulase Negative Staphylococcus|Coagulase Positive Staphylococcus|Beta-haemolytic Streptococcus|unknown Gram negatives|unknown Gram positives|CoNS|CoPS|no MO|Gram negative|Gram positive|Bacteria|Fungi|Protozoa|biogroup|biotype|vegetative|group|Group)") if (sum(x_tobetranslated, na.rm = TRUE) == 0) { return(x) } # only translate the ones that need translation x[x_tobetranslated] <- case_when( # German language == "de" ~ x[x_tobetranslated] %>% gsub("Coagulase Negative Staphylococcus","Koagulase-negative Staphylococcus", ., fixed = TRUE) %>% gsub("Coagulase Positive Staphylococcus","Koagulase-positive Staphylococcus", ., fixed = TRUE) %>% gsub("Beta-haemolytic Streptococcus", "Beta-h\u00e4molytischer Streptococcus", ., fixed = TRUE) %>% gsub("unknown Gram negatives", "unbekannte Gramnegativen", ., fixed = TRUE) %>% gsub("unknown Gram positives", "unbekannte Grampositiven", ., fixed = TRUE) %>% gsub("(CoNS)", "(KNS)", ., fixed = TRUE) %>% gsub("(CoPS)", "(KPS)", ., fixed = TRUE) %>% gsub("(no MO)", "(kein MO)", ., fixed = TRUE) %>% gsub("Gram negative", "Gramnegativ", ., fixed = TRUE) %>% gsub("Gram positive", "Grampositiv", ., fixed = TRUE) %>% gsub("Bacteria", "Bakterien", ., fixed = TRUE) %>% gsub("Fungi", "Hefen/Pilze", ., fixed = TRUE) %>% gsub("Protozoa", "Protozoen", ., fixed = TRUE) %>% gsub("biogroup", "Biogruppe", ., fixed = TRUE) %>% gsub("biotype", "Biotyp", ., fixed = TRUE) %>% gsub("vegetative", "vegetativ", ., fixed = TRUE) %>% gsub("([([ ]*?)group", "\\1Gruppe", .) %>% gsub("([([ ]*?)Group", "\\1Gruppe", .) %>% iconv(to = "UTF-8"), # Dutch language == "nl" ~ x[x_tobetranslated] %>% gsub("Coagulase Negative Staphylococcus","Coagulase-negatieve Staphylococcus", ., fixed = TRUE) %>% gsub("Coagulase Positive Staphylococcus","Coagulase-positieve Staphylococcus", ., fixed = TRUE) %>% gsub("Beta-haemolytic Streptococcus", "Beta-hemolytische Streptococcus", ., fixed = TRUE) %>% gsub("unknown Gram negatives", "onbekende Gram-negatieven", ., fixed = TRUE) %>% gsub("unknown Gram positives", "onbekende Gram-positieven", ., fixed = TRUE) %>% gsub("(no MO)", "(geen MO)", ., fixed = TRUE) %>% gsub("(CoNS)", "(CNS)", ., fixed = TRUE) %>% gsub("(CoPS)", "(CPS)", ., fixed = TRUE) %>% gsub("Gram negative", "Gram-negatief", ., fixed = TRUE) %>% gsub("Gram positive", "Gram-positief", ., fixed = TRUE) %>% gsub("Bacteria", "Bacteri\u00ebn", ., fixed = TRUE) %>% gsub("Fungi", "Schimmels/gisten", ., fixed = TRUE) %>% gsub("Protozoa", "protozo\u00ebn", ., fixed = TRUE) %>% gsub("biogroup", "biogroep", ., fixed = TRUE) %>% # gsub("biotype", "biotype", ., fixed = TRUE) %>% gsub("vegetative", "vegetatief", ., fixed = TRUE) %>% gsub("([([ ]*?)group", "\\1groep", .) %>% gsub("([([ ]*?)Group", "\\1Groep", .) %>% iconv(to = "UTF-8"), # Spanish language == "es" ~ x[x_tobetranslated] %>% gsub("Coagulase Negative Staphylococcus","Staphylococcus coagulasa negativo", ., fixed = TRUE) %>% gsub("Coagulase Positive Staphylococcus","Staphylococcus coagulasa positivo", ., fixed = TRUE) %>% gsub("Beta-haemolytic Streptococcus", "Streptococcus Beta-hemol\u00edtico", ., fixed = TRUE) %>% gsub("unknown Gram negatives", "Gram negativos desconocidos", ., fixed = TRUE) %>% gsub("unknown Gram positives", "Gram positivos desconocidos", ., fixed = TRUE) %>% gsub("(no MO)", "(sin MO)", ., fixed = TRUE) %>% gsub("Gram negative", "Gram negativo", ., fixed = TRUE) %>% gsub("Gram positive", "Gram positivo", ., fixed = TRUE) %>% gsub("Bacteria", "Bacterias", ., fixed = TRUE) %>% gsub("Fungi", "Hongos", ., fixed = TRUE) %>% gsub("Protozoa", "Protozoarios", ., fixed = TRUE) %>% gsub("biogroup", "biogrupo", ., fixed = TRUE) %>% gsub("biotype", "biotipo", ., fixed = TRUE) %>% gsub("vegetative", "vegetativo", ., fixed = TRUE) %>% gsub("([([ ]*?)group", "\\1grupo", .) %>% gsub("([([ ]*?)Group", "\\1Grupo", .) %>% iconv(to = "UTF-8"), # Italian language == "it" ~ x[x_tobetranslated] %>% gsub("Coagulase Negative Staphylococcus","Staphylococcus negativo coagulasi", ., fixed = TRUE) %>% gsub("Coagulase Positive Staphylococcus","Staphylococcus positivo coagulasi", ., fixed = TRUE) %>% gsub("Beta-haemolytic Streptococcus", "Streptococcus Beta-emolitico", ., fixed = TRUE) %>% gsub("unknown Gram negatives", "Gram negativi sconosciuti", ., fixed = TRUE) %>% gsub("unknown Gram positives", "Gram positivi sconosciuti", ., fixed = TRUE) %>% gsub("(no MO)", "(non MO)", ., fixed = TRUE) %>% gsub("Gram negative", "Gram negativo", ., fixed = TRUE) %>% gsub("Gram positive", "Gram positivo", ., fixed = TRUE) %>% gsub("Bacteria", "Batteri", ., fixed = TRUE) %>% gsub("Fungi", "Fungo", ., fixed = TRUE) %>% gsub("Protozoa", "Protozoi", ., fixed = TRUE) %>% gsub("biogroup", "biogruppo", ., fixed = TRUE) %>% gsub("biotype", "biotipo", ., fixed = TRUE) %>% gsub("vegetative", "vegetativo", ., fixed = TRUE) %>% gsub("([([ ]*?)group", "\\1gruppo", .) %>% gsub("([([ ]*?)Group", "\\1Gruppo", .), # French language == "fr" ~ x[x_tobetranslated] %>% gsub("Coagulase Negative Staphylococcus","Staphylococcus \u00e0 coagulase n\u00e9gative", ., fixed = TRUE) %>% gsub("Coagulase Positive Staphylococcus","Staphylococcus \u00e0 coagulase positif", ., fixed = TRUE) %>% gsub("Beta-haemolytic Streptococcus", "Streptococcus B\u00eata-h\u00e9molytique", ., fixed = TRUE) %>% gsub("unknown Gram negatives", "Gram n\u00e9gatifs inconnus", ., fixed = TRUE) %>% gsub("unknown Gram positives", "Gram positifs inconnus", ., fixed = TRUE) %>% gsub("(no MO)", "(pas MO)", ., fixed = TRUE) %>% gsub("Gram negative", "Gram n\u00e9gatif", ., fixed = TRUE) %>% gsub("Gram positive", "Gram positif", ., fixed = TRUE) %>% gsub("Bacteria", "Bact\u00e9ries", ., fixed = TRUE) %>% gsub("Fungi", "Champignons", ., fixed = TRUE) %>% gsub("Protozoa", "Protozoaires", ., fixed = TRUE) %>% gsub("biogroup", "biogroupe", ., fixed = TRUE) %>% # gsub("biotype", "biotype", ., fixed = TRUE) %>% gsub("vegetative", "v\u00e9g\u00e9tatif", ., fixed = TRUE) %>% gsub("([([ ]*?)group", "\\1groupe", .) %>% gsub("([([ ]*?)Group", "\\1Groupe", .) %>% iconv(to = "UTF-8"), # Portuguese language == "pt" ~ x[x_tobetranslated] %>% gsub("Coagulase Negative Staphylococcus","Staphylococcus coagulase negativo", ., fixed = TRUE) %>% gsub("Coagulase Positive Staphylococcus","Staphylococcus coagulase positivo", ., fixed = TRUE) %>% gsub("Beta-haemolytic Streptococcus", "Streptococcus Beta-hemol\u00edtico", ., fixed = TRUE) %>% gsub("unknown Gram negatives", "Gram negativos desconhecidos", ., fixed = TRUE) %>% gsub("unknown Gram positives", "Gram positivos desconhecidos", ., fixed = TRUE) %>% gsub("(no MO)", "(sem MO)", ., fixed = TRUE) %>% gsub("Gram negative", "Gram negativo", ., fixed = TRUE) %>% gsub("Gram positive", "Gram positivo", ., fixed = TRUE) %>% gsub("Bacteria", "Bact\u00e9rias", ., fixed = TRUE) %>% gsub("Fungi", "Fungos", ., fixed = TRUE) %>% gsub("Protozoa", "Protozo\u00e1rios", ., fixed = TRUE) %>% gsub("biogroup", "biogrupo", ., fixed = TRUE) %>% gsub("biotype", "bi\u00f3tipo", ., fixed = TRUE) %>% gsub("vegetative", "vegetativo", ., fixed = TRUE) %>% gsub("([([ ]*?)group", "\\1grupo", .) %>% gsub("([([ ]*?)Group", "\\1Grupo", .) %>% iconv(to = "UTF-8")) x } mo_validate <- function(x, property, ...) { dots <- list(...) Becker <- dots$Becker if (is.null(Becker)) { Becker <- FALSE } Lancefield <- dots$Lancefield if (is.null(Lancefield)) { Lancefield <- FALSE } if (!"AMR" %in% base::.packages()) { library("AMR") # check onLoad() in R/zzz.R: data tables are created there. } if (!all(x %in% microorganismsDT[[property]]) | Becker %in% c(TRUE, "all") | Lancefield %in% c(TRUE, "all")) { exec_as.mo(x, property = property, ...) } else { if (property == "mo") { return(structure(x, class = "mo")) } else { return(x) } } }