Mercurial > repos > artbio > repenrich
view RepEnrich.py @ 15:2e3d976e7d5d draft default tip
planemo upload for repository https://github.com/ARTbio/tools-artbio/tree/main/tools/repenrich commit 03183e29f807ec33548016a7c4144f52720b7b9e
author | artbio |
---|---|
date | Sun, 21 Apr 2024 09:44:51 +0000 |
parents | bf866bedd4b4 |
children |
line wrap: on
line source
import argparse import csv import shlex import subprocess import sys from collections import defaultdict from concurrent.futures import ProcessPoolExecutor parser = argparse.ArgumentParser(description=''' Repenrich aligns reads to Repeat Elements pseudogenomes\ and counts aligned reads. RepEnrich_setup must be run\ before its use''') parser.add_argument('--annotation_file', action='store', metavar='annotation_file', help='RepeatMasker.org annotation file for your\ organism. The file may be downloaded from\ RepeatMasker.org. E.g. hg19_repeatmasker.txt') parser.add_argument('--alignment_bam', action='store', metavar='alignment_bam', help='Bam alignments of unique mapper reads.') parser.add_argument('--fastqfile', action='store', metavar='fastqfile', help='File of fastq reads mapping to multiple\ locations. Example: /data/multimap.fastq') parser.add_argument('--fastqfile2', action='store', dest='fastqfile2', metavar='fastqfile2', default='', help='fastqfile #2 when using paired-end option.\ Default none') parser.add_argument('--cpus', action='store', dest='cpus', metavar='cpus', default="1", type=int, help='Number of CPUs. The more cpus the\ faster RepEnrich performs. Default: "1"') args = parser.parse_args() # parameters annotation_file = args.annotation_file unique_mapper_bam = args.alignment_bam fastqfile_1 = args.fastqfile fastqfile_2 = args.fastqfile2 cpus = args.cpus # Change if simple repeats are differently annotated in your organism simple_repeat = "Simple_repeat" if args.fastqfile2: paired_end = True else: paired_end = False def starts_with_numerical(list): try: if len(list) == 0: return False int(list[0]) return True except ValueError: return False # define a text importer for .out/.txt format of repbase def import_text(filename, separator): csv.field_size_limit(sys.maxsize) file = csv.reader(open(filename), delimiter=separator, skipinitialspace=True) return [line for line in file if starts_with_numerical(line)] def run_bowtie(args): metagenome, fastqfile = args b_opt = "-k 1 -p 1 --quiet" command = shlex.split(f"bowtie {b_opt} -x {metagenome} {fastqfile}") bowtie_align = subprocess.run(command, check=True, capture_output=True, text=True).stdout bowtie_align = bowtie_align.rstrip('\r\n').split('\n') readlist = [metagenome] if paired_end: for line in bowtie_align: readlist.append(line.split("/")[0]) else: for line in bowtie_align: readlist.append(line.split("\t")[0]) return readlist # set a reference repeat list for the script repeat_list = [listline[9].translate( str.maketrans( '()/', '___')) for listline in import_text(annotation_file, ' ')] repeat_list = sorted(list(set(repeat_list))) # unique mapper counting cmd = f"bedtools bamtobed -i {unique_mapper_bam} | \ bedtools coverage -b stdin -a repnames.bed" p = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE) bedtools_counts = p.communicate()[0].decode().rstrip('\r\n').split('\n') # parse bedtools output counts = defaultdict(int) # key: repeat names, value: unique mapper counts sumofrepeatreads = 0 for line in bedtools_counts: line = line.split('\t') counts[line[3]] += int(line[4]) sumofrepeatreads += int(line[4]) print(f"Identified {sumofrepeatreads} unique reads that mapped to repeats.") # multimapper parsing if not paired_end: args_list = [(metagenome, fastqfile_1) for metagenome in repeat_list] with ProcessPoolExecutor(max_workers=cpus) as executor: results = executor.map(run_bowtie, args_list) else: args_list = [(metagenome, fastqfile_1) for metagenome in repeat_list] args_list.extend([(metagenome, fastqfile_2) for metagenome in repeat_list]) with ProcessPoolExecutor(max_workers=cpus) as executor: results = executor.map(run_bowtie, args_list) # Aggregate results (avoiding race conditions) metagenome_reads = defaultdict(list) # repeat_name: list of multimap reads for result in results: metagenome_reads[result[0]] += result[1:] for name in metagenome_reads: # read are only once in list metagenome_reads[name] = list(set(metagenome_reads[name])) # remove "no read" instances metagenome_reads[name] = [read for read in metagenome_reads[name] if read != ""] # implement repeats_by_reads from the inverse dictionnary metagenome_reads repeats_by_reads = defaultdict(list) # readids: list of repeats names for repname in metagenome_reads: for read in metagenome_reads[repname]: repeats_by_reads[read].append(repname) for repname in repeats_by_reads: repeats_by_reads[repname] = list(set(repeats_by_reads[repname])) # 3 dictionnaries and 1 pointer variable to be populated fractionalcounts = defaultdict(float) familyfractionalcounts = defaultdict(float) classfractionalcounts = defaultdict(float) sumofrepeatreads = 0 # Update counts dictionnary with sets of repeats (was "subfamilies") # matched by multimappers for repeat_set in repeats_by_reads.values(): repeat_set_string = ','.join(repeat_set) counts[repeat_set_string] += 1 sumofrepeatreads += 1 print(f'Identified more {sumofrepeatreads} mutimapper repeat reads') # Populate fractionalcounts for key, count in counts.items(): key_list = key.split(',') for i in key_list: fractionalcounts[i] += count / len(key_list) # build repeat_ref for easy access to rep class and rep families repeat_ref = defaultdict(dict) repeats = import_text(annotation_file, ' ') for repeat in repeats: repeat_name = repeat[9].translate(str.maketrans('()/', '___')) try: repclass = repeat[10].split('/')[0] repfamily = repeat[10].split('/')[1] except IndexError: repclass, repfamily = repeat[10], repeat[10] repeat_ref[repeat_name]['class'] = repclass repeat_ref[repeat_name]['family'] = repfamily # Populate classfractionalcounts and familyfractionalcounts for key, value in fractionalcounts.items(): classfractionalcounts[repeat_ref[key]['class']] += value familyfractionalcounts[repeat_ref[key]['family']] += value # print class-, family- and fraction-repeats counts to files with open("class_fraction_counts.tsv", 'w') as fout: for key in sorted(classfractionalcounts): fout.write(f"{key}\t{classfractionalcounts[key]}\n") with open("family_fraction_counts.tsv", 'w') as fout: for key in sorted(familyfractionalcounts): fout.write(f"{key}\t{familyfractionalcounts[key]}\n") with open("fraction_counts.tsv", 'w') as fout: for key in sorted(fractionalcounts): fout.write(f"{key}\t{repeat_ref[key]['class']}\t" f"{repeat_ref[key]['family']}\t" f"{fractionalcounts[key]}\n")