mirror of
https://github.com/MillironX/taxprofiler.git
synced 2024-11-25 13:59:55 +00:00
94 lines
4 KiB
Text
94 lines
4 KiB
Text
//
|
|
// Check input samplesheet and get read channels
|
|
//
|
|
|
|
include { UNTAR } from '../../modules/nf-core/untar/main'
|
|
|
|
workflow DB_CHECK {
|
|
take:
|
|
dbsheet // file: /path/to/dbsheet.csv
|
|
|
|
main:
|
|
ch_versions = Channel.empty()
|
|
ch_dbs_for_untar = Channel.empty()
|
|
ch_final_dbs = Channel.empty()
|
|
|
|
// Special check to check _between_ rows, for which we must group rows together
|
|
// Note: this will run in parallel to within-row validity, but we can assume this will run faster thus will fail first
|
|
Channel.fromPath(dbsheet)
|
|
.splitCsv ( header:true, sep:',' )
|
|
.map {[it.tool, it.db_name] }
|
|
.groupTuple()
|
|
.map {
|
|
tool, db_name ->
|
|
def unique_names = db_name.unique(false)
|
|
if ( unique_names.size() < db_name.size() ) exit 1, "[nf-core/taxprofiler] ERROR: Each database for a tool must have a unique name, duplicated detected. Tool: ${tool}, Database name: ${unique_names}"
|
|
}
|
|
|
|
// Normal checks for within-row validity, so can be moved to separate functions
|
|
parsed_samplesheet = Channel.fromPath(dbsheet)
|
|
.splitCsv ( header:true, sep:',' )
|
|
.map {
|
|
validate_db_rows(it)
|
|
create_db_channels(it)
|
|
}
|
|
|
|
ch_dbs_for_untar = parsed_samplesheet
|
|
.branch {
|
|
untar: it[1].toString().endsWith(".tar.gz")
|
|
skip: true
|
|
}
|
|
|
|
// Filter the channel to untar only those databases for tools that are selected to be run by the user.
|
|
ch_input_untar = ch_dbs_for_untar.untar
|
|
.filter {
|
|
params["run_${it[0]['tool']}"]
|
|
}
|
|
|
|
UNTAR (ch_input_untar)
|
|
ch_versions = ch_versions.mix(UNTAR.out.versions.first())
|
|
ch_final_dbs = ch_dbs_for_untar.skip.mix( UNTAR.out.untar )
|
|
|
|
emit:
|
|
dbs = ch_final_dbs // channel: [ val(meta), [ db ] ]
|
|
versions = ch_versions // channel: [ versions.yml ]
|
|
}
|
|
|
|
def validate_db_rows(LinkedHashMap row){
|
|
|
|
// check minimum number of columns
|
|
if (row.size() < 4) exit 1, "[nf-core/taxprofiler] ERROR: Invalid database input sheet - malformed row (e.g. missing column). See documentation for more information. Error in: ${row}"
|
|
|
|
// all columns there
|
|
def expected_headers = ['tool', 'db_name', 'db_params', 'db_path']
|
|
if ( !row.keySet().containsAll(expected_headers) ) exit 1, "[nf-core/taxprofiler] ERROR: Invalid database input sheet - malformed column names. Please check input TSV. Column names should be: ${expected_keys.join(", ")}"
|
|
|
|
// valid tools specified// TIFNISIH LIST
|
|
def expected_tools = [ "bracken", "centrifuge", "diamond", "kaiju", "kraken2", "krakenuniq", "malt", "metaphlan3", "motus" ]
|
|
if ( !expected_tools.contains(row.tool) ) exit 1, "[nf-core/taxprofiler] ERROR: Invalid tool name. Please see documentation for all supported profilers. Error in: ${row}"
|
|
|
|
// detect quotes in params
|
|
if ( row.db_params.contains('"') ) exit 1, "[nf-core/taxprofiler] ERROR: Invalid database db_params entry. No quotes allowed. Error in: ${row}"
|
|
if ( row.db_params.contains("'") ) exit 1, "[nf-core/taxprofiler] ERROR: Invalid database db_params entry. No quotes allowed. Error in: ${row}"
|
|
|
|
// check if any form of bracken params, that it must have `;`
|
|
if ( row.tool == 'bracken' && row.db_params && !row.db_params.contains(";") ) exit 1, "[nf-core/taxprofiler] ERROR: Invalid database db_params entry. Bracken requires a semi-colon if passing parameter. Error in: ${row}"
|
|
|
|
}
|
|
|
|
def create_db_channels(LinkedHashMap row) {
|
|
def meta = [:]
|
|
meta.tool = row.tool
|
|
meta.db_name = row.db_name
|
|
meta.db_params = row.db_params
|
|
|
|
def array = []
|
|
if (!file(row.db_path, type: 'dir').exists()) {
|
|
exit 1, "ERROR: Please check input samplesheet -> database path could not be found!\n${row.db_path}"
|
|
}
|
|
array = [ meta, file(row.db_path) ]
|
|
|
|
return array
|
|
}
|
|
|
|
|