Nothing
# ------------------------------------------------------------------------------
#' @title Annotate table with Gene Symbol and Name based on Uniprot ID(s)
#'
#' @description Annotate gene name and symbol based on uniprot ids. It will
#' take the column from your data.frame specified by the `columnid` argument,
#' search for the gene symbol, name, and entrez based on the species (`species`
#' argument) and merge the information back to the input data.frame
#' @param x (data.frame) to be annotated (or file path and name)
#' @param columnid (char) The column with the uniprotkb ids
#' @param species (char) The species name. Check `?artmsMapUniprot2Entrez`
#' to find out more about supported species.
#' @param verbose (logical) `TRUE` (default) shows function messages
#' @return (data.frame) with two new columns: `Gene` and `Protein.name`
#' @keywords annotation, uniprot
#' @examples
#' # This example adds annotations to the evidence file available in
#' # artMS, based on the column 'Proteins'.
#'
#' evidence_anno <- artmsAnnotationUniprot(x = artms_data_ph_evidence,
#' columnid = 'Proteins',
#' species = 'human')
#' @export
artmsAnnotationUniprot <- function(x,
columnid,
species,
verbose = TRUE) {
if(any(missing(x) |
missing(columnid) |
missing(species)))
stop("Missed (one or many) required argument(s)
Please, check the help of this function to find out more")
if(!is.character(columnid)) stop("Argument 'columnid' must be a character")
if(!is.character(species)) stop("Argument <species> must be a character")
x <- .artms_checkIfFile(x)
# Deal with isoforms UNIPROT-1 etc
if(!columnid %in% colnames(x)){
stop("The column id <",columnid,"> was not found in this dataset")
}
x$EntryRoot <- x[[columnid]]
# Check first if it is uniprot:
isUniprot <- TRUE
if( length(x$EntryRoot[grep("\\w{2}_\\d{1,}\\.\\d{1,}", x$EntryRoot)]) > 100 ){
message("\nWarning!
It looks like this dataset is using RefSeq as protein ID.
RefSeq is not (yet) supported for gene annotation.\n")
isUniprot <- FALSE
}
x$EntryRoot <- gsub("(\\w+)(-)(\\d+)", "\\1", x$EntryRoot)
# Deal with protein groups and PTMs: select the first ID
# 1. strsplit based on ;
x$EntryRoot <- unlist(lapply(x$EntryRoot, function(x) unlist(strsplit(x, "\\;"))[1]))
# 2. now based on PTM: either STYK
x$EntryRoot <- unlist(lapply(x$EntryRoot, function(x) unlist(strsplit(x, "_[S|T|Y|K]"))[1]))
theUniprots <- as.character(unique(x$EntryRoot))
if( isFALSE(artmsIsSpeciesSupported(species = species)) | isFALSE(isUniprot) ){
if(isFALSE(artmsIsSpeciesSupported(species = species)))
message("---(-) Species ", species," not supported: no info about proteins will be provided")
keepSearchName <- paste0(columnid, "Copy")
y <- artmsChangeColumnName(x, columnid, keepSearchName)
y <- artmsChangeColumnName(y, "EntryRoot", "Protein")
y$Gene <- y$Protein
y$ProteinName <- NA
y$EntrezID <- NA
send_back <- y[c("Protein", "Gene", "ProteinName", "EntrezID")]
send_back <- unique(send_back)
y <- subset(y, select = -c(Gene, ProteinName, EntrezID))
send_back <- merge(send_back, y, by = "Protein")
return(send_back)
}else{
preload <- artmsMapUniprot2Entrez(uniprotkb = theUniprots,
species = species)
dc_merged <-merge(x,
preload,
by.x = "EntryRoot",
by.y = "UNIPROT",
all.x = TRUE)
# Move Gene name to the left:
gene_first <- preload[, c("SYMBOL", "UNIPROT", "GENENAME", "ENTREZID")]
dc_merged <- subset(dc_merged, select = -c(SYMBOL, GENENAME, ENTREZID))
send_back <-
merge(
gene_first,
dc_merged,
by.x = "UNIPROT",
by.y = "EntryRoot",
all.y = TRUE
)
keepSearchName <- paste0(columnid,"Ref")
send_back <- artmsChangeColumnName(send_back, columnid, keepSearchName)
names(send_back)[grep("^UNIPROT$", names(send_back))] <- "Protein"
names(send_back)[grep("^SYMBOL$", names(send_back))] <- "Gene"
names(send_back)[grep("^GENENAME$", names(send_back))] <- "ProteinName"
names(send_back)[grep("^ENTREZID$", names(send_back))] <- "EntrezID"
# Some uniprot entries might not have yet a gene name,
# which will be an empty value. Replace with Entry name:
send_back$Gene[which(send_back$Gene == "")] <- NA
send_back$Gene[is.na(send_back$Gene)] <- send_back$Protein[is.na(send_back$Gene)]
return(send_back)
}
}
# ------------------------------------------------------------------------------
# @title Select the entry
#
# @description Selet the Uniprot Entry ID from a full Uniprot id. For example:
#
# From `sp|P55011|S12A2_HUMAN` will select `P55011`
#
# @param x (vector) of protein ids. If the id is a uniprot full entry, it will
# extract the ENTRY id
# @return (vector) with only the Entry ID (if found)
# @keywords internal, Protein ID, Uniprot, Entry
.artms_selectEntryFromFullUniprot <- function(x){
isolateIt <- unlist(strsplit(x, ";"))
replaceIt <- as.character(sapply(isolateIt, function(y) gsub("(^sp|tr)(\\|)(.*)(\\|.*)", "\\3", y)))
if(length(replaceIt) > 1){
replaceIt <- paste(replaceIt, collapse = ";")
}
return(replaceIt)
}
# ------------------------------------------------------------------------------
#' @title Leave only the Entry ID from a typical full Uniprot IDs in a
#' given column
#'
#' @description Downloading a Reference Uniprot fasta database includes several
#' Uniprot IDs for every protein. If the regular expression available in
#' Maxquant is not activated, the full id will be used in the Proteins,
#' Lead Protein, and Leading Razor Protein columns. This script leaves only the
#' Entry ID.
#'
#' For example, values in a Protein column like this:
#'
#' `sp|P12345|Entry_name;sp|P54321|Entry_name2`
#'
#' will be replace by
#'
#' `P12345;P54321``
#'
#' @param x (data.frame) that contains the `columnid`
#' @param columnid (char) Column name with the full uniprot ids
#' @return (data.frame) with only Entry IDs.
#'
#' @keywords annotation, ids
#' @examples
#' # Example of data frame with full uniprot ids and sequences
#' p <- c("sp|A6NIE6|RN3P2_HUMAN;sp|Q9NYV6|RRN3_HUMAN",
#' "sp|A7E2V4|ZSWM8_HUMAN",
#' "sp|A5A6H4|ROA1_PANTR;sp|P09651|ROA1_HUMAN;sp|Q32P51|RA1L2_HUMAN",
#' "sp|A0FGR8|ESYT2_HUMAN")
#' s <- c("ALENDFFNSPPRK", "GWGSPGRPK", "SSGPYGGGGQYFAK", "VLVALASEELAK")
#' evidence <- data.frame(Proteins = p, Sequences = s, stringsAsFactors = FALSE)
#'
#' # Replace the Proteins column with only Entry ids
#' evidence <- artmsLeaveOnlyUniprotEntryID(x = evidence, columnid = "Proteins")
#' @export
artmsLeaveOnlyUniprotEntryID <- function(x, columnid){
if(!(is.data.frame(x) | is.data.table(x))){
stop("<x> must be a data.frame or data.table")
}
if(!is.vector(columnid)){
stop("<columnid> must be a valid column name")
}
if(!(columnid %in% colnames(x))){
stop("The <columnid> is not a column of <x>. Check the name again")
}
x[[columnid]] <- unlist(lapply(x[[columnid]],
function(z) sapply(z, .artms_selectEntryFromFullUniprot)))
return(x)
}
# ------------------------------------------------------------------------------
#' @title Map GENE SYMBOL, NAME, AND ENTREZID to a vector of Uniprot IDS
#'
#' @description Map GENE SYMBOL, NAME, AND ENTREZID to a vector of Uniprot IDS
#' @param uniprotkb (vector) Vector of UniprotKB IDs
#' @param species (char) The species name. Species currently supported
#' as part of artMS: check `?artmsIsSpeciesSupported()` to find out the
#' list of supported species`
#' @return (data.frame) with ENTREZID and GENENAMES mapped on UniprotKB ids
#' @keywords annotation, ids
#' @examples
#' # Load an example with human proteins
#' exampleID <- c("Q6P996", "B1N8M6")
#' artmsMapUniprot2Entrez(uniprotkb = exampleID,
#' species = "HUMAN")
#'
#' @export
artmsMapUniprot2Entrez <- function(uniprotkb,
species) {
if(any(missing(uniprotkb) |
missing(species)))
stop("Missed (one or many) required argument(s)
Please, check the help of this function to find out more")
if(!is.vector(uniprotkb)) stop("Argument <uniprotkb> is not a vector")
if(!is.character(species)) stop("Argument <species> is not a vector")
if(isFALSE(artmsIsSpeciesSupported(species = species))){
stop("Species ", species," not supported")
}else{
thePack <- artmsIsSpeciesSupported(species = species)
}
suppressMessages(
mappings <-
AnnotationDbi::select(
eval(as.symbol(thePack)),
uniprotkb,
c("UNIPROT", "SYMBOL",
"GENENAME", "ENTREZID"),
keytype = "UNIPROT"
)
)
mappings <- unique(mappings)
# It migth come with 1 uniprot to many gene names:
# take the first one, which should be the main gene name
mappings <- mappings[!duplicated(mappings$UNIPROT),]
return(mappings)
}
# ------------------------------------------------------------------------------
#' @title Check if a species is supported and available
#'
#' @description Given a species name, it checkes whether is supported, and
#' if supported, check whether the annotation package is installed.
#' @param species (char) The species name. Species currently supported
#' as part of artMS:
#' - HUMAN
#' - MOUSE
#'
#' And the following species can be used as well, but the user needs to
#' install the corresponding org.db package:
#' - ANOPHELES (`install.packages(org.Ag.eg.db)`)
#' - BOVINE (`install.packages(org.Bt.eg.db)`)
#' - WORM (`install.packages(org.Ce.eg.db)`)
#' - CANINE (`install.packages(org.Cf.eg.db)`)
#' - FLY (`install.packages(org.Dm.eg.db)`)
#' - ZEBRAFISH (`install.packages(org.Dr.eg.db)`)
#' - CHICKEN (`install.packages(org.Gg.eg.db)`)
#' - RHESUS (`install.packages(org.Mmu.eg.db)`)
#' - CHIMP (`install.packages(org.Pt.eg.db)`)
#' - RAT (`install.packages(org.Rn.eg.db)`)
#' - YEAST (`install.packages(org.Sc.sgd.db)`)
#' - PIG (`install.packages(org.Ss.eg.db)`)
#' - XENOPUS (`install.packages(org.Xl.eg.db)`)
#' @param verbose (logical) `TRUE` (default) shows function messages
#' @return (string) Name of the package for the given species
#' @keywords annotation, species
#' @examples
#' # Should return TRUE
#' artmsIsSpeciesSupported(species = "HUMAN")
#' artmsIsSpeciesSupported(species = "CHIMP")
#' @export
artmsIsSpeciesSupported <- function(species,
verbose = TRUE) {
if(any(missing(species)))
stop("Missed (one or many) required argument(s)
Please, check the help of this function to find out more")
if(!is.character(species)) stop("Argument <species> is not a vector")
species <- toupper(species)
LongName <- c(
"ANOPHELES",
"BOVINE",
"WORM",
"CANINE",
"FLY",
"ZEBRAFISH",
"CHICKEN",
"HUMAN",
"MOUSE",
"RHESUS",
"CHIMP",
"RAT",
"YEAST",
"PIG",
"XENOPUS")
PackageName <- c(
"org.Ag.eg.db",
"org.Bt.eg.db",
"org.Ce.eg.db",
"org.Cf.eg.db",
"org.Dm.eg.db",
"org.Dr.eg.db",
"org.Gg.eg.db",
"org.Hs.eg.db",
"org.Mm.eg.db",
"org.Mmu.eg.db",
"org.Pt.eg.db",
"org.Rn.eg.db",
"org.Sc.sgd.db",
"org.Ss.eg.db",
"org.Xl.eg.db")
OrgDB <- data.frame("LongName" = LongName,
"PackageName" = PackageName,
stringsAsFactors = FALSE)
if(species %in% OrgDB$LongName){
thePack <- OrgDB$PackageName[which(OrgDB$LongName == species)]
if( !(thePack %in% rownames(installed.packages())) ){
if(verbose) message("---(-) The package <",thePack,"> is not installed in your system.
Just run: install.packages('", thePack,"') and try again")
return(FALSE)
} else{
return(thePack)
}
}else{
return(FALSE)
}
}
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.