Nothing
#' @title RNASeqRParam_Bam
#'
#' @description An S4 class for checking and storing RNA-Seq workflow
#' parameters starting with BAM files.
#'
#'
#' @slot os.type 'linux' or 'osx'. The operating system type.
#' @slot python.variable A list storing python environment.
#' \code{(check.answer, python.version)}
#' @slot python.2to3 Logical value whether \code{2to3} command is available
#' on the workstation.
#' @slot path.prefix Path prefix of 'gene_data/', 'RNASeq_bin/',
#' 'RNASeq_results/', 'Rscript/' and 'Rscript_out/' directories.
#' @slot input.path.prefix Path prefix of 'input_files/' directory,
#' @slot genome.name Variable of genome name defined in this RNA-Seq workflow
#' (ex. \code{genome.name}.fa, \code{genome.name}.gtf).
#' @slot sample.pattern Regular expression of paired-end fastq.gz files under
#' 'input_files/raw_bam'. Expression not includes \code{_[1,2].fastq.gz}.
#' @slot independent.variable Independent variable for the biological.
#' experiment design of two-group RNA-Seq workflow.
#' @slot case.group Group name of the case group.
#' @slot control.group Group name of the control group.
#'
#' @name RNASeqRParam_Bam-class
#'
#' @rdname RNASeqRParam_Bam-class
#'
#' @exportClass RNASeqRParam_Bam
#' @author Kuan-Hao Chao
#' @examples
#' data(yeast)
#' "@@"(yeast, os.type)
#' "@@"(yeast, python.variable)
#' "@@"(yeast, python.2to3)
#' "@@"(yeast, path.prefix)
#' "@@"(yeast, input.path.prefix)
#' "@@"(yeast, genome.name)
#' "@@"(yeast, sample.pattern)
#' "@@"(yeast, independent.variable)
#' "@@"(yeast, case.group)
#' "@@"(yeast, control.group)
setClass("RNASeqRParam_Bam",
representation(
os.type = "character",
python.variable = "list",
python.2to3 = "logical",
path.prefix = "character",
input.path.prefix = "character",
genome.name = "character",
sample.pattern = "character",
independent.variable = "character",
case.group = "character",
control.group = "character"
)
)
#' @title RNASeqR_Bam
#' @description Constructor function for RNASeqRParam_Bam objects
#'
#' @name RNASeqRParam_Bam-constructor
#'
#' @param path.prefix Path prefix of 'gene_data/', 'RNASeq_bin/',
#' 'RNASeq_results/', 'Rscript/' and 'Rscript_out/' directories.
#' @param input.path.prefix Path prefix of 'input_files/' directory.
#' @param genome.name variable of genome name defined in this RNA-Seq workflow
#' (ex. \code{genome.name}.fa, \code{genome.name}.gtf).
#' @param sample.pattern Regular expression of paired-end fastq.gz files under
#' 'input_files/raw_bam'. Expression not includes \code{_[1,2].fastq.gz}.
#' @param independent.variable Independent variable for the biological
#' experiment design of two-group RNA-Seq workflow.
#' @param case.group Group name of the case group.
#' @param control.group Group name of the control group.
#'
#' @return an object of class \code{RNASeqRParam_Bam}
#'
#' @author Kuan-Hao Chao
#'
#' @rdname RNASeqRParam_Bam-constructor
#'
#' @export
#' @author Kuan-Hao Chao
#' @examples
#' input_files.path <- system.file("extdata/", package = "RNASeqRData")
#' rnaseq_result.path <- tempdir(check = TRUE)
#' \dontrun{
#' exp <- RNASeqRParam_Bam(path.prefix = rnaseq_result.path,
#' input.path.prefix = input_files.path,
#' genome.name = "Saccharomyces_cerevisiae_XV_Ensembl",
#' sample.pattern = "SRR[0-9]*_XV",
#' independent.variable = "state",
#' case.group = "60mins_ID20_amphotericin_B",
#' control.group = "60mins_ID20_control")
#' }
RNASeqRParam_Bam <- function(path.prefix = NA,
input.path.prefix = NA,
genome.name = NA,
sample.pattern = NA,
independent.variable = NA,
case.group = NA,
control.group = NA) {
# check input parameters
CheckInputParamNa(path.prefix,
input.path.prefix,
genome.name,
sample.pattern,
independent.variable,
case.group,
control.group,
"")
# 1. check operating system
characters.os.type <- CheckOperatingSystem()
# 2. check python version
python.version.list <- CheckPython()
bool.python.avail <- python.version.list$check.answer
two.to.three.result <- Check2to3()
# 3. check validity of path.prefix
bool.prefix.path <- CheckPrefixPath(path.prefix = path.prefix)
if (bool.prefix.path){
# add '/' to the path.prefix
if (substr(path.prefix, nchar(path.prefix), nchar(path.prefix)) != "/") {
path.prefix <- paste0(path.prefix, "/")
}
}
# 4. check input.path.prefix
bool.input.path.prefix <- CheckInputPrefixPath(input.path.prefix)
if (bool.input.path.prefix){
# add '/' to the path.prefix
if (substr(input.path.prefix,
nchar(input.path.prefix),
nchar(input.path.prefix)) != "/") {
input.path.prefix <- paste0(input.path.prefix, "/")
}
}
# 5. check 'phenodata'
bool.phenodata <- CheckPhenodata_Bam(input.path.prefix,
genome.name,
sample.pattern,
independent.variable)
# 6. check 'input_files/' necessary files with 'genome.name', 'sample.pattern'
input.dir.files.list <- CheckInputDirFiles_Bam(input.path.prefix,
genome.name,
sample.pattern)
bool.input.dir.files <- input.dir.files.list
# This determine whether to run 'CreateHisat2Index'
# 7. check 'case.group' and 'control.group'
bool.control.control.group <- CheckCaseControlGroup(input.path.prefix,
independent.variable,
case.group, control.group)
if ( (characters.os.type == "linux" || characters.os.type == "osx") &&
bool.python.avail && bool.prefix.path && bool.input.path.prefix &&
bool.input.dir.files && bool.phenodata && bool.control.control.group) {
message("\n**************************************\n")
message("************** Success! **************\n")
message("**************************************\n")
new("RNASeqRParam_Bam",
os.type = characters.os.type,
python.variable = python.version.list,
python.2to3 = two.to.three.result,
path.prefix = path.prefix,
input.path.prefix = input.path.prefix,
genome.name = genome.name,
sample.pattern = sample.pattern,
independent.variable = independent.variable,
case.group = case.group,
control.group = control.group)
}
}
# inner function : check input.path.prefix
CheckInputDirFiles_Bam <- function(input.path.prefix, genome.name, sample.pattern) {
message("************** Checking hierarchy of '",
input.path.prefix, "input_files/' ************\n")
# only check whether exist
gtf.file <- file.exists(paste0(input.path.prefix,
"input_files/",
genome.name, ".gtf"))
# check exist and rules
raw.bam.dir <- dir.exists(paste0(input.path.prefix,
"input_files/raw_bam/"))
# check exist and rules
phenodata.file <- file.exists(paste0(input.path.prefix,
"input_files/phenodata.csv"))
# check whether sample pattern matches the file names~
if (raw.bam.dir) {
raw.bam <- list.files(path = paste0(input.path.prefix,
"input_files/raw_bam/"),
pattern = "*.bam$",
all.files = FALSE,
full.names = FALSE,
recursive = FALSE,
ignore.case = FALSE)
raw.bam.number <- length(raw.bam)
if (raw.bam.number != 0) {
message("(\u2714) : '*.bam' is in 'input_files'\n")
} else {
message("(\u231B) : '", input.path.prefix,
"input_files/raw_bam/XXX.bam' is not exit\n")
}
} else {
message("(\u231B) : '", input.path.prefix,
"input_files/raw_bam/XXX.bam' is not exit\n")
}
if (!isTRUE(gtf.file)) {
grab.gtf.file <- Sys.glob(file.path(path = paste0(input.path.prefix,
"input_files"), "*.gtf"))
message("(\u2718) : '", genome.name,
".gtf (user input)' and '", grab.gtf.file,
" (find in directory)' ", " are mismatched.\n")
} else {
message("(\u2714) : '", genome.name,
".gtf'", " is in 'input_files'\n")
}
if (!isTRUE(phenodata.file)) {
message("(\u2718) : '", "phenodata.csv is missing.\n")
} else {
message("(\u2714) : '", "phenodata.csv is in 'input_files'\n")
}
if (gtf.file && phenodata.file) {
message("\n(\u2714) : '", input.path.prefix,
"input_files/", "' is valid !\n")
message("\n")
return.value <- TRUE
return(return.value)
} else {
stop("'input_files/' checking ERROR")
}
}
# inner function : check validity of phenodata
CheckPhenodata_Bam <- function(input.path.prefix,
genome.name,
sample.pattern,
independent.variable) {
# have to sort the column !! and sort them in the correct order
message("************** Checking phenodata ************\n")
pheno_data <- read.csv(paste0(input.path.prefix, "input_files/phenodata.csv"))
# Covert all column to character
pheno_data <- data.frame(lapply(pheno_data, as.character),
stringsAsFactors = FALSE)
# Check 'ids' is in the 'phenodata.csv'
if (!("ids" %in% colnames(pheno_data))) {
message("(\u2718) : 'ids' can't find in the ",
"column of phenodata.csv.\n\n")
stop("'ids' invalid ERROR")
}
# "id" : must be distinct, same as input_files raw reads name !
message("\u25B6 Checking whether \"raw_bam files\" ",
"matches \"'ids' of phenodata.csv\" \n")
raw.bam <- list.files(path = paste0(input.path.prefix,
"input_files/raw_bam/"),
pattern = sample.pattern,
all.files = FALSE,
full.names = FALSE,
recursive = FALSE,
ignore.case = FALSE)
extract.bam.sample.names <- unique(gsub(".bam", "", raw.bam))
bool.length <- length(extract.bam.sample.names) == length(pheno_data$ids)
bool.identical <- identical(sort(extract.bam.sample.names),
sort(pheno_data$ids))
if (!bool.length || !bool.identical) {
message("(\u2718) : 'ids' column doesn't match the smaple_id in ",
"'input_files/raw_bam'. Please check the file and ",
"'input_files/raw_bam' directory.\n" )
stop("'ids' mismatch ERROR")
}
ids.list <- paste(sort(extract.bam.sample.names), collapse = ", ")
message("(\u2714) : Column 'ids' of phenodata.csv is valid. \n")
message(" \u25CF sample ids are : \"", ids.list, "\"\n")
# Check again independent.variable in the list
if (!(independent.variable %in% colnames(pheno_data))) {
message("(\u2718) : 'independent.variable' : '",
independent.variable,
"' can't find in the column of phenodata.csv.\n\n")
stop("'independent.variable' invalid ERROR")
}
message("(\u2714) : 'independent.variable' : '",
independent.variable,
"' is in the column of phenodata.csv. \n\n")
message("\u25B6 Checking whether '",
independent.variable,
"' is a two-group 'independent.variable' ...\n")
length.independent.variable <- length(table(pheno_data[independent.variable]))
if (!(length.independent.variable == 2)) {
message("(\u2718) : 'independent.variable' : '",
independent.variable, "' is a ",
length.independent.variable,
"-group 'independent.variable'. Not 2-group.\n")
message(" groups that found : ",
paste0(names(table(pheno_data[independent.variable])),
collapse = ", "), "\n")
stop("'independent.variable' none-two-group ERROR")
}
message("(\u2714) : Column 'independent.variable' : '",
independent.variable, "' of phenodata.csv is valid. \n")
message(" \u25CF 'independent.variable' : '",
independent.variable, "'\n\n")
return(TRUE)
}
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.