Mercurial > repos > prog > lcmsmatching
view KeggEntry.R @ 2:20d69a062da3 draft
planemo upload for repository https://github.com/workflow4metabolomics/lcmsmatching.git commit d4048accde6bdfd5b3e14f5394902d38991854f8
author | prog |
---|---|
date | Thu, 02 Mar 2017 08:55:00 -0500 |
parents | |
children |
line wrap: on
line source
##################### # CLASS DECLARATION # ##################### KeggEntry <- methods::setRefClass("KeggEntry", contains = 'BiodbEntry') ########### # FACTORY # ########### createKeggEntryFromTxt <- function(contents, drop = TRUE) { entries <- list() # Define fields regex regex <- character() regex[[BIODB.NAME]] <- "^NAME\\s+([^,;]+)" regex[[BIODB.CHEBI.ID]] <- "^\\s+ChEBI:\\s+(\\S+)" regex[[BIODB.LIPIDMAPS.ID]] <- "^\\s+LIPIDMAPS:\\s+(\\S+)" for (text in contents) { # Create instance entry <- KeggEntry$new() lines <- strsplit(text, "\n") for (s in lines[[1]]) { # Test generic regex parsed <- FALSE for (field in names(regex)) { g <- stringr::str_match(s, regex[[field]]) if ( ! is.na(g[1,1])) { entry$setField(field, g[1,2]) parsed <- TRUE break } } if (parsed) next # ACCESSION { # ENZYME ID g <- stringr::str_match(s, "^ENTRY\\s+EC\\s+(\\S+)") if ( ! is.na(g[1,1])){ entry$setField(BIODB.ACCESSION, paste('ec', g[1,2], sep = ':')) # ENTRY ID }else { g <- stringr::str_match(s, "^ENTRY\\s+(\\S+)\\s+Compound") if ( ! is.na(g[1,1])){ entry$setField(BIODB.ACCESSION, paste('cpd', g[1,2], sep = ':')) # OTHER ID }else { g <- stringr::str_match(s, "^ENTRY\\s+(\\S+)") if ( ! is.na(g[1,1])) entry$setField(BIODB.ACCESSION, g[1,2]) } } # ORGANISM g <- stringr::str_match(s, "^ORGANISM\\s+(\\S+)") if ( ! is.na(g[1,1])) entry$setField(BIODB.ACCESSION, paste(g[1,2], entry$getField(BIODB.ACCESSION), sep = ':')) } } entries <- c(entries, entry) } # Replace elements with no accession id by NULL entries <- lapply(entries, function(x) if (is.na(x$getField(BIODB.ACCESSION))) NULL else x) # If the input was a single element, then output a single object if (drop && length(contents) == 1) entries <- entries[[1]] return(entries) }