mirror of https://github.com/msberends/AMR.git
268 lines
10 KiB
R
268 lines
10 KiB
R
# ==================================================================== #
|
|
# TITLE #
|
|
# Antimicrobial Resistance (AMR) Analysis #
|
|
# #
|
|
# SOURCE #
|
|
# https://github.com/msberends/AMR #
|
|
# #
|
|
# LICENCE #
|
|
# (c) 2018-2020 Berends MS, Luz CF et al. #
|
|
# #
|
|
# 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 more info: https://msberends.github.io/AMR. #
|
|
# ==================================================================== #
|
|
|
|
#' User-defined reference data set for microorganisms
|
|
#'
|
|
#' @description These functions can be used to predefine your own reference to be used in [as.mo()] and consequently all `mo_*` functions like [mo_genus()] and [mo_gramstain()].
|
|
#'
|
|
#' This is **the fastest way** to have your organisation (or analysis) specific codes picked up and translated by this package.
|
|
#' @inheritSection lifecycle Stable lifecycle
|
|
#' @param path location of your reference file, see Details. Can be `""`, `NULL` or `FALSE` to delete the reference file.
|
|
#' @rdname mo_source
|
|
#' @name mo_source
|
|
#' @aliases set_mo_source get_mo_source
|
|
#' @details The reference file can be a text file seperated with commas (CSV) or tabs or pipes, an Excel file (either 'xls' or 'xlsx' format) or an R object file (extension '.rds'). To use an Excel file, you need to have the `readxl` package installed.
|
|
#'
|
|
#' [set_mo_source()] will check the file for validity: it must be a [`data.frame`], must have a column named `"mo"` which contains values from [`microorganisms$mo`][microorganisms] and must have a reference column with your own defined values. If all tests pass, [set_mo_source()] will read the file into R and export it to `"~/.mo_source.rds"`. This compressed data file will then be used at default for MO determination (function [as.mo()] and consequently all `mo_*` functions like [mo_genus()] and [mo_gramstain()]). The location of the original file will be saved as option with `options(mo_source = path)`. Its timestamp will be saved with `options(mo_source_datetime = ...)`.
|
|
#'
|
|
#' [get_mo_source()] will return the data set by reading `"~/.mo_source.rds"` with [readRDS()]. If the original file has changed (the file defined with `path`), it will call [set_mo_source()] to update the data file automatically.
|
|
#'
|
|
#' Reading an Excel file (`.xlsx`) with only one row has a size of 8-9 kB. The compressed file created with [set_mo_source()] will then have a size of 0.1 kB and can be read by [get_mo_source()] in only a couple of microseconds (millionths of a second).
|
|
#'
|
|
#' @section How to setup:
|
|
#'
|
|
#' Imagine this data on a sheet of an Excel file (mo codes were looked up in the [microorganisms] data set). The first column contains the organisation specific codes, the second column contains an MO code from this package:
|
|
#'
|
|
#' ```
|
|
#' | A | B |
|
|
#' --|--------------------|--------------|
|
|
#' 1 | Organisation XYZ | mo |
|
|
#' 2 | lab_mo_ecoli | B_ESCHR_COLI |
|
|
#' 3 | lab_mo_kpneumoniae | B_KLBSL_PNMN |
|
|
#' 4 | | |
|
|
#' ```
|
|
#'
|
|
#' We save it as `"home/me/ourcodes.xlsx"`. Now we have to set it as a source:
|
|
#'
|
|
#' ```
|
|
#' set_mo_source("home/me/ourcodes.xlsx")
|
|
#' #> NOTE: Created mo_source file '~/.mo_source.rds' from 'home/me/ourcodes.xlsx'
|
|
#' #> (columns "Organisation XYZ" and "mo")
|
|
#' ```
|
|
#'
|
|
#' It has now created a file `"~/.mo_source.rds"` with the contents of our Excel file. Only the first column with foreign values and the 'mo' column will be kept when creating the RDS file.
|
|
#'
|
|
#' And now we can use it in our functions:
|
|
#'
|
|
#' ```
|
|
#' as.mo("lab_mo_ecoli")
|
|
#' #> [1] B_ESCHR_COLI
|
|
#'
|
|
#' mo_genus("lab_mo_kpneumoniae")
|
|
#' #> [1] "Klebsiella"
|
|
#'
|
|
#' # other input values still work too
|
|
#' as.mo(c("Escherichia coli", "E. coli", "lab_mo_ecoli"))
|
|
#' #> [1] B_ESCHR_COLI B_ESCHR_COLI B_ESCHR_COLI
|
|
#' ```
|
|
#'
|
|
#' If we edit the Excel file by, let's say, adding row 4 like this:
|
|
#'
|
|
#' ```
|
|
#' | A | B |
|
|
#' --|--------------------|--------------|
|
|
#' 1 | Organisation XYZ | mo |
|
|
#' 2 | lab_mo_ecoli | B_ESCHR_COLI |
|
|
#' 3 | lab_mo_kpneumoniae | B_KLBSL_PNMN |
|
|
#' 4 | lab_Staph_aureus | B_STPHY_AURS |
|
|
#' 5 | | |
|
|
#' ```
|
|
#'
|
|
#' ...any new usage of an MO function in this package will update your data file:
|
|
#'
|
|
#' ```
|
|
#' as.mo("lab_mo_ecoli")
|
|
#' #> NOTE: Updated mo_source file '~/.mo_source.rds' from 'home/me/ourcodes.xlsx'
|
|
#' #> (columns "Organisation XYZ" and "mo")
|
|
#' #> [1] B_ESCHR_COLI
|
|
#'
|
|
#' mo_genus("lab_Staph_aureus")
|
|
#' #> [1] "Staphylococcus"
|
|
#' ```
|
|
#'
|
|
#' To delete the reference data file, just use `""`, `NULL` or `FALSE` as input for [set_mo_source()]:
|
|
#'
|
|
#' ```
|
|
#' set_mo_source(NULL)
|
|
#' # Removed mo_source file '~/.mo_source.rds'.
|
|
#' ```
|
|
#'
|
|
#' If the original Excel file is moved or deleted, the mo_source file will be removed upon the next use of [as.mo()]. If the mo_source file is manually deleted (i.e. without using [set_mo_source()]), the references to the mo_source file will be removed upon the next use of [as.mo()].
|
|
#' @export
|
|
#' @inheritSection AMR Read more on our website!
|
|
set_mo_source <- function(path) {
|
|
|
|
file_location <- path.expand("~/mo_source.rds")
|
|
|
|
stop_ifnot(length(path) == 1, "`path` must be of length 1")
|
|
|
|
if (is.null(path) || path %in% c(FALSE, "")) {
|
|
options(mo_source = NULL)
|
|
options(mo_source_timestamp = NULL)
|
|
if (file.exists(file_location)) {
|
|
unlink(file_location)
|
|
message(font_red(paste0("Removed mo_source file '", font_bold(file_location), "'")))
|
|
}
|
|
return(invisible())
|
|
}
|
|
|
|
stop_ifnot(file.exists(path),
|
|
"file not found: ", path)
|
|
|
|
if (path %like% "[.]rds$") {
|
|
df <- readRDS(path)
|
|
|
|
} else if (path %like% "[.]xlsx?$") {
|
|
# is Excel file (old or new)
|
|
read_excel <- import_fn("read_excel", "readxl")
|
|
df <- read_excel(path)
|
|
|
|
} else if (path %like% "[.]tsv$") {
|
|
df <- utils::read.table(header = TRUE, sep = "\t", stringsAsFactors = FALSE)
|
|
|
|
} else {
|
|
# try comma first
|
|
try(
|
|
df <- utils::read.table(header = TRUE, sep = ",", stringsAsFactors = FALSE),
|
|
silent = TRUE)
|
|
if (!mo_source_isvalid(df, stop_on_error = FALSE)) {
|
|
# try tab
|
|
try(
|
|
df <- utils::read.table(header = TRUE, sep = "\t", stringsAsFactors = FALSE),
|
|
silent = TRUE)
|
|
}
|
|
if (!mo_source_isvalid(df, stop_on_error = FALSE)) {
|
|
# try pipe
|
|
try(
|
|
df <- utils::read.table(header = TRUE, sep = "|", stringsAsFactors = FALSE),
|
|
silent = TRUE)
|
|
}
|
|
}
|
|
|
|
# check integrity
|
|
mo_source_isvalid(df)
|
|
|
|
df <- subset(df, !is.na(mo))
|
|
|
|
# keep only first two columns, second must be mo
|
|
if (colnames(df)[1] == "mo") {
|
|
df <- df[, c(colnames(df)[2], "mo")]
|
|
} else {
|
|
df <- df[, c(colnames(df)[1], "mo")]
|
|
}
|
|
|
|
df <- as.data.frame(df, stringAsFactors = FALSE)
|
|
|
|
# success
|
|
if (file.exists(file_location)) {
|
|
action <- "Updated"
|
|
} else {
|
|
action <- "Created"
|
|
}
|
|
saveRDS(df, file_location)
|
|
options(mo_source = path)
|
|
options(mo_source_timestamp = as.character(file.info(path)$mtime))
|
|
message(font_blue(paste0("NOTE: ",
|
|
action, " mo_source file '", font_bold(file_location), "'",
|
|
" from '", font_bold(path), "'",
|
|
'\n (columns "', colnames(df)[1], '" and "', colnames(df)[2], '")')))
|
|
}
|
|
|
|
#' @rdname mo_source
|
|
#' @export
|
|
get_mo_source <- function() {
|
|
if (is.null(getOption("mo_source", NULL))) {
|
|
return(NULL)
|
|
}
|
|
|
|
if (!file.exists(path.expand("~/mo_source.rds"))) {
|
|
options(mo_source = NULL)
|
|
options(mo_source_timestamp = NULL)
|
|
message(font_blue("NOTE: Removed references to deleted mo_source file (see ?mo_source)"))
|
|
return(NULL)
|
|
}
|
|
|
|
old_time <- as.POSIXct(getOption("mo_source_timestamp"))
|
|
new_time <- as.POSIXct(as.character(file.info(getOption("mo_source", ""))$mtime))
|
|
|
|
if (is.na(new_time)) {
|
|
# source file was deleted, remove reference too
|
|
set_mo_source("")
|
|
return(NULL)
|
|
}
|
|
if (new_time != old_time) {
|
|
# set updated source
|
|
set_mo_source(getOption("mo_source"))
|
|
}
|
|
file_location <- path.expand("~/mo_source.rds")
|
|
readRDS(file_location)
|
|
}
|
|
|
|
mo_source_isvalid <- function(x, refer_to_name = "`reference_df`", stop_on_error = TRUE) {
|
|
|
|
check_dataset_integrity()
|
|
|
|
if (deparse(substitute(x)) == "get_mo_source()") {
|
|
return(TRUE)
|
|
}
|
|
if (identical(x, get_mo_source())) {
|
|
return(TRUE)
|
|
}
|
|
if (is.null(x)) {
|
|
if (stop_on_error == TRUE) {
|
|
stop(refer_to_name, " cannot be NULL", call. = FALSE)
|
|
} else {
|
|
return(FALSE)
|
|
}
|
|
}
|
|
if (!is.data.frame(x)) {
|
|
if (stop_on_error == TRUE) {
|
|
stop(refer_to_name, " must be a data.frame", call. = FALSE)
|
|
} else {
|
|
return(FALSE)
|
|
}
|
|
}
|
|
if (!"mo" %in% colnames(x)) {
|
|
if (stop_on_error == TRUE) {
|
|
stop(refer_to_name, " must contain a column 'mo'", call. = FALSE)
|
|
} else {
|
|
return(FALSE)
|
|
}
|
|
}
|
|
if (!all(x$mo %in% c("", microorganisms$mo, microorganisms.translation$mo_old), na.rm = TRUE)) {
|
|
if (stop_on_error == TRUE) {
|
|
invalid <- x[which(!x$mo %in% c("", microorganisms$mo, microorganisms.translation$mo_old)), , drop = FALSE]
|
|
if (nrow(invalid) > 1) {
|
|
plural <- "s"
|
|
} else {
|
|
plural <- ""
|
|
}
|
|
stop("Value", plural, " ", paste0("'", invalid[, 1, drop = TRUE], "'", collapse = ", "),
|
|
" found in ", tolower(refer_to_name),
|
|
", but with invalid microorganism code", plural, " ", paste0("'", invalid$mo, "'", collapse = ", "),
|
|
call. = FALSE)
|
|
} else {
|
|
return(FALSE)
|
|
}
|
|
}
|
|
TRUE
|
|
}
|