Mercurial > repos > drosofff > repenrich
view RepEnrich.py @ 2:aed130b47d36 draft
planemo upload for repository https://github.com/ARTbio/tools-artbio/tree/master/tools/repenrich commit 5bd2ca88fb31e5e836f41bc3c70b9d559bcdb52f
author | drosofff |
---|---|
date | Mon, 29 May 2017 17:08:36 -0400 |
parents | 1435d142041b |
children |
line wrap: on
line source
#!/usr/bin/env python import argparse import csv import numpy import os import shlex import shutil import subprocess import sys parser = argparse.ArgumentParser(description='Part II: Conducting the alignments to the psuedogenomes. Before doing this step you will require 1) a bamfile of the unique alignments with index 2) a fastq file of the reads mapping to more than one location. These files can be obtained using the following bowtie options [EXAMPLE: bowtie -S -m 1 --max multimap.fastq mm9 mate1_reads.fastq] Once you have the unique alignment bamfile and the reads mapping to more than one location in a fastq file you can run this step. EXAMPLE: python master_output.py /users/nneretti/data/annotation/hg19/hg19_repeatmasker.txt /users/nneretti/datasets/repeatmapping/POL3/Pol3_human/HeLa_InputChIPseq_Rep1 HeLa_InputChIPseq_Rep1 /users/nneretti/data/annotation/hg19/setup_folder HeLa_InputChIPseq_Rep1_multimap.fastq HeLa_InputChIPseq_Rep1.bam') parser.add_argument('--version', action='version', version='%(prog)s 0.1') parser.add_argument('annotation_file', action= 'store', metavar='annotation_file', help='List RepeatMasker.org annotation file for your organism. The file may be downloaded from the RepeatMasker.org website. Example: /data/annotation/hg19/hg19_repeatmasker.txt') parser.add_argument('outputfolder', action= 'store', metavar='outputfolder', help='List folder to contain results. Example: /outputfolder') parser.add_argument('outputprefix', action= 'store', metavar='outputprefix', help='Enter prefix name for data. Example: HeLa_InputChIPseq_Rep1') parser.add_argument('setup_folder', action= 'store', metavar='setup_folder', help='List folder that contains the repeat element psuedogenomes. Example /data/annotation/hg19/setup_folder') parser.add_argument('fastqfile', action= 'store', metavar='fastqfile', help='Enter file for the fastq reads that map to multiple locations. Example /data/multimap.fastq') parser.add_argument('alignment_bam', action= 'store', metavar='alignment_bam', help='Enter bamfile output for reads that map uniquely. Example /bamfiles/old.bam') parser.add_argument('--pairedend', action= 'store', dest='pairedend', default= 'FALSE', help='Designate this option for paired-end sequencing. Default FALSE change to TRUE') parser.add_argument('--collapserepeat', action= 'store', dest='collapserepeat', metavar='collapserepeat', default= 'Simple_repeat', help='Designate this option to generate a collapsed repeat type. Uncollapsed output is generated in addition to collapsed repeat type. Simple_repeat is default to simplify downstream analysis. You can change the default to another repeat name to collapse a seperate specific repeat instead or if the name of Simple_repeat is different for your organism. Default Simple_repeat') parser.add_argument('--fastqfile2', action= 'store', dest='fastqfile2', metavar='fastqfile2', default= 'none', help='Enter fastqfile2 when using paired-end option. Default none') parser.add_argument('--cpus', action= 'store', dest='cpus', metavar='cpus', default= "1", type=int, help='Enter available cpus per node. The more cpus the faster RepEnrich performs. RepEnrich is designed to only work on one node. Default: "1"') parser.add_argument('--allcountmethod', action= 'store', dest='allcountmethod', metavar='allcountmethod', default= "FALSE", help='By default the pipeline only outputs the fraction count method. Consdidered to be the best way to count multimapped reads. Changing this option will include the unique count method, a conservative count, and the total count method, a liberal counting strategy. Our evaluation of simulated data indicated fraction counting is best. Default = FALSE, change to TRUE') parser.add_argument('--is_bed', action= 'store', dest='is_bed', metavar='is_bed', default= 'FALSE', help='Is the annotation file a bed file. This is also a compatible format. The file needs to be a tab seperated bed with optional fields. Ex. format chr\tstart\tend\tName_element\tclass\tfamily. The class and family should identical to name_element if not applicable. Default FALSE change to TRUE') args = parser.parse_args() # parameters annotation_file = args.annotation_file outputfolder = args.outputfolder outputfile_prefix = args.outputprefix setup_folder = args.setup_folder repeat_bed = setup_folder + os.path.sep + 'repnames.bed' unique_mapper_bam = args.alignment_bam fastqfile_1 = args.fastqfile fastqfile_2 = args.fastqfile2 cpus = args.cpus b_opt = "-k1 -p " +str(1) +" --quiet" simple_repeat = args.collapserepeat paired_end = args.pairedend allcountmethod = args.allcountmethod is_bed = args.is_bed ################################################################################ # check that the programs we need are available try: subprocess.call(shlex.split("coverageBed -h"), stdout=open(os.devnull, 'wb'), stderr=open(os.devnull, 'wb')) subprocess.call(shlex.split("bowtie --version"), stdout=open(os.devnull, 'wb'), stderr=open(os.devnull, 'wb')) except OSError: print ("Error: Bowtie or BEDTools not loaded") raise ################################################################################ # define a csv reader that reads space deliminated files print ('Preparing for analysis using RepEnrich...') csv.field_size_limit(sys.maxsize) def import_text(filename, separator): for line in csv.reader(open(filename), delimiter=separator, skipinitialspace=True): if line: yield line ################################################################################ # build dictionaries to convert repclass and rep families' if is_bed == "FALSE": repeatclass = {} repeatfamily = {} fin = import_text(annotation_file, ' ') x = 0 for line in fin: if x>2: classfamily =[] classfamily = line[10].split(os.path.sep) line9 = line[9].replace("(","_").replace(")","_").replace("/","_") repeatclass[line9] = classfamily[0] if len(classfamily) == 2: repeatfamily[line9] = classfamily[1] else: repeatfamily[line9] = classfamily[0] x +=1 if is_bed == "TRUE": repeatclass = {} repeatfamily = {} fin = open(annotation_file, 'r') for line in fin: line=line.strip('\n') line=line.split('\t') theclass =line[4] thefamily = line[5] line3 = line[3].replace("(","_").replace(")","_").replace("/","_") repeatclass[line3] = theclass repeatfamily[line3] = thefamily fin.close() ################################################################################ # build list of repeats initializing dictionaries for downstream analysis' fin = import_text(setup_folder + os.path.sep + 'repgenomes_key.txt', '\t') repeat_key ={} rev_repeat_key ={} repeat_list = [] reptotalcounts = {} classfractionalcounts = {} familyfractionalcounts = {} classtotalcounts = {} familytotalcounts = {} reptotalcounts_simple = {} fractionalcounts = {} i = 0 for line in fin: reptotalcounts[line[0]] = 0 fractionalcounts[line[0]] = 0 if line[0] in repeatclass: classtotalcounts[repeatclass[line[0]]] = 0 classfractionalcounts[repeatclass[line[0]]] = 0 if line[0] in repeatfamily: familytotalcounts[repeatfamily[line[0]]] = 0 familyfractionalcounts[repeatfamily[line[0]]] = 0 if line[0] in repeatfamily: if repeatfamily[line[0]] == simple_repeat: reptotalcounts_simple[simple_repeat] = 0 else: reptotalcounts_simple[line[0]] = 0 repeat_list.append(line[0]) repeat_key[line[0]] = int(line[1]) rev_repeat_key[int(line[1])] = line[0] fin.close() ################################################################################ # map the repeats to the psuedogenomes: if not os.path.exists(outputfolder): os.mkdir(outputfolder) ################################################################################ # Conduct the regions sorting print ('Conducting region sorting on unique mapping reads....') fileout= outputfolder + os.path.sep + outputfile_prefix + '_regionsorter.txt' with open(fileout, 'w') as stdout: command = shlex.split("coverageBed -abam " +unique_mapper_bam+" -b " +setup_folder + os.path.sep + 'repnames.bed') p = subprocess.Popen(command, stdout=stdout) p.communicate() stdout.close() filein = open(outputfolder + os.path.sep + outputfile_prefix + '_regionsorter.txt','r') counts = {} sumofrepeatreads=0 for line in filein: line= line.split('\t') if not str(repeat_key[line[3]]) in counts: counts[str(repeat_key[line[3]])]=0 counts[str(repeat_key[line[3]])]+=int(line[4]) sumofrepeatreads+=int(line[4]) print ('Identified ' + str(sumofrepeatreads) + 'unique reads that mapped to repeats.') ################################################################################ if paired_end == 'TRUE': if not os.path.exists(outputfolder + os.path.sep + 'pair1_bowtie'): os.mkdir(outputfolder + os.path.sep + 'pair1_bowtie') if not os.path.exists(outputfolder + os.path.sep + 'pair2_bowtie'): os.mkdir(outputfolder + os.path.sep + 'pair2_bowtie') folder_pair1 = outputfolder + os.path.sep + 'pair1_bowtie' folder_pair2 = outputfolder + os.path.sep + 'pair2_bowtie' ################################################################################ print ("Processing repeat psuedogenomes...") ps = [] psb= [] ticker= 0 for metagenome in repeat_list: metagenomepath = setup_folder + os.path.sep + metagenome file1=folder_pair1 + os.path.sep + metagenome + '.bowtie' file2 =folder_pair2 + os.path.sep + metagenome + '.bowtie' with open(file1, 'w') as stdout: command = shlex.split("bowtie " + b_opt + " " + metagenomepath + " " + fastqfile_1) p = subprocess.Popen(command,stdout=stdout) with open(file2, 'w') as stdout: command = shlex.split("bowtie " + b_opt + " " + metagenomepath + " " + fastqfile_2) pp = subprocess.Popen(command,stdout=stdout) ps.append(p) ticker +=1 psb.append(pp) ticker +=1 if ticker == cpus: for p in ps: p.communicate() for p in psb: p.communicate() ticker = 0 psb =[] ps = [] if len(ps) > 0: for p in ps: p.communicate() stdout.close() ################################################################################ # combine the output from both read pairs: print ('sorting and combining the output for both read pairs...') if not os.path.exists(outputfolder + os.path.sep + 'sorted_bowtie'): os.mkdir(outputfolder + os.path.sep + 'sorted_bowtie') sorted_bowtie = outputfolder + os.path.sep + 'sorted_bowtie' for metagenome in repeat_list: file1 = folder_pair1 + os.path.sep + metagenome + '.bowtie' file2 = folder_pair2 + os.path.sep + metagenome + '.bowtie' fileout= sorted_bowtie + os.path.sep + metagenome + '.bowtie' with open(fileout, 'w') as stdout: p1 = subprocess.Popen(['cat',file1,file2], stdout = subprocess.PIPE) p2 = subprocess.Popen(['cut', '-f1',"-d "], stdin = p1.stdout, stdout = subprocess.PIPE) p3 = subprocess.Popen(['cut', '-f1', "-d/"], stdin = p2.stdout, stdout = subprocess.PIPE) p4 = subprocess.Popen(['sort'], stdin=p3.stdout, stdout = subprocess.PIPE) p5 = subprocess.Popen(['uniq'], stdin=p4.stdout, stdout = stdout) p5.communicate() stdout.close() print ('completed ...') ################################################################################ if paired_end == 'FALSE': if not os.path.exists(outputfolder + os.path.sep + 'pair1_bowtie'): os.mkdir(outputfolder + os.path.sep + 'pair1_bowtie') folder_pair1 = outputfolder + os.path.sep + 'pair1_bowtie' ################################################################################ ps = [] ticker= 0 print ("Processing repeat psuedogenomes...") for metagenome in repeat_list: metagenomepath = setup_folder + os.path.sep + metagenome file1=folder_pair1 + os.path.sep + metagenome + '.bowtie' with open(file1, 'w') as stdout: command = shlex.split("bowtie " + b_opt + " " + metagenomepath + " " + fastqfile_1) p = subprocess.Popen(command,stdout=stdout) ps.append(p) ticker +=1 if ticker == cpus: for p in ps: p.communicate() ticker = 0 ps = [] if len(ps) > 0: for p in ps: p.communicate() stdout.close() ################################################################################ # combine the output from both read pairs: print ('Sorting and combining the output for both read pairs....') if not os.path.exists(outputfolder + os.path.sep + 'sorted_bowtie'): os.mkdir(outputfolder + os.path.sep + 'sorted_bowtie') sorted_bowtie = outputfolder + os.path.sep + 'sorted_bowtie' for metagenome in repeat_list: file1 = folder_pair1 + os.path.sep + metagenome + '.bowtie' fileout= sorted_bowtie + os.path.sep + metagenome + '.bowtie' with open(fileout, 'w') as stdout: p1 = subprocess.Popen(['cat',file1], stdout = subprocess.PIPE) p2 = subprocess.Popen(['cut', '-f1'], stdin = p1.stdout, stdout = subprocess.PIPE) p3 = subprocess.Popen(['cut', '-f1', "-d/"], stdin = p2.stdout, stdout = subprocess.PIPE) p4 = subprocess.Popen(['sort'], stdin = p3.stdout,stdout = subprocess.PIPE) p5 = subprocess.Popen(['uniq'], stdin = p4.stdout,stdout = stdout) p5.communicate() stdout.close() print ('completed ...') ################################################################################ # build a file of repeat keys for all reads print ('Writing and processing intermediate files...') sorted_bowtie = outputfolder + os.path.sep + 'sorted_bowtie' readid = {} sumofrepeatreads=0 for rep in repeat_list: for data in import_text(sorted_bowtie + os.path.sep + rep + '.bowtie', '\t'): readid[data[0]] = '' for rep in repeat_list: for data in import_text(sorted_bowtie + os.path.sep + rep + '.bowtie', '\t'): readid[data[0]]+=str(repeat_key[rep]) + str(',') for subfamilies in readid.values(): if not subfamilies in counts: counts[subfamilies]=0 counts[subfamilies] +=1 sumofrepeatreads+=1 del readid print ('Identified ' + str(sumofrepeatreads) + ' reads that mapped to repeats for unique and multimappers.') ################################################################################ print ("Conducting final calculations...") # build a converter to numeric label for repeat and yield a combined list of repnames seperated by backslash def convert(x): x = x.strip(',') x = x.split(',') global repname repname = "" for i in x: repname = repname + os.path.sep + rev_repeat_key[int(i)] # building the total counts for repeat element enrichment... for x in counts.keys(): count= counts[x] x = x.strip(',') x = x.split(',') for i in x: reptotalcounts[rev_repeat_key[int(i)]] += int(count) # building the fractional counts for repeat element enrichment... for x in counts.keys(): count= counts[x] x = x.strip(',') x = x.split(',') splits = len(x) for i in x: fractionalcounts[rev_repeat_key[int(i)]] += float(numpy.divide(float(count),float(splits))) # building categorized table of repeat element enrichment... repcounts = {} repcounts['other'] = 0 for key in counts.keys(): convert(key) repcounts[repname] = counts[key] # building the total counts for class enrichment... for key in reptotalcounts.keys(): classtotalcounts[repeatclass[key]] += reptotalcounts[key] # building total counts for family enrichment... for key in reptotalcounts.keys(): familytotalcounts[repeatfamily[key]] += reptotalcounts[key] # building unique counts table' repcounts2 = {} for rep in repeat_list: if "/" +rep in repcounts: repcounts2[rep] = repcounts["/" +rep] else: repcounts2[rep] = 0 # building the fractionalcounts counts for class enrichment... for key in fractionalcounts.keys(): classfractionalcounts[repeatclass[key]] += fractionalcounts[key] # building fractional counts for family enrichment... for key in fractionalcounts.keys(): familyfractionalcounts[repeatfamily[key]] += fractionalcounts[key] ################################################################################ print ('Writing final output and removing intermediate files...') # print output to file of the categorized counts and total overlapping counts: if allcountmethod == "TRUE": fout1 = open(outputfolder + os.path.sep + outputfile_prefix + '_total_counts.txt' , 'w') for key in reptotalcounts.keys(): fout1.write(str(key) + '\t' + repeatclass[key] + '\t' + repeatfamily[key] + '\t' + str(reptotalcounts[key]) + '\n') fout2 = open(outputfolder + os.path.sep + outputfile_prefix + '_class_total_counts.txt' , 'w') for key in classtotalcounts.keys(): fout2.write(str(key) + '\t' + str(classtotalcounts[key]) + '\n') fout3 = open(outputfolder + os.path.sep + outputfile_prefix + '_family_total_counts.txt' , 'w') for key in familytotalcounts.keys(): fout3.write(str(key) + '\t' + str(familytotalcounts[key]) + '\n') fout4 = open(outputfolder + os.path.sep + outputfile_prefix + '_unique_counts.txt' , 'w') for key in repcounts2.keys(): fout4.write(str(key) + '\t' + repeatclass[key] + '\t' + repeatfamily[key] + '\t' + str(repcounts2[key]) + '\n') fout5 = open(outputfolder + os.path.sep + outputfile_prefix + '_class_fraction_counts.txt' , 'w') for key in classfractionalcounts.keys(): fout5.write(str(key) + '\t' + str(classfractionalcounts[key]) + '\n') fout6 = open(outputfolder + os.path.sep + outputfile_prefix + '_family_fraction_counts.txt' , 'w') for key in familyfractionalcounts.keys(): fout6.write(str(key) + '\t' + str(familyfractionalcounts[key]) + '\n') fout7 = open(outputfolder + os.path.sep + outputfile_prefix + '_fraction_counts.txt' , 'w') for key in fractionalcounts.keys(): fout7.write(str(key) + '\t' + repeatclass[key] + '\t' + repeatfamily[key] + '\t' + str(int(fractionalcounts[key])) + '\n') fout1.close() fout2.close() fout3.close() fout4.close() fout5.close() fout6.close() fout7.close() else: fout1 = open(outputfolder + os.path.sep + outputfile_prefix + '_class_fraction_counts.txt' , 'w') for key in classfractionalcounts.keys(): fout1.write(str(key) + '\t' + str(classfractionalcounts[key]) + '\n') fout2 = open(outputfolder + os.path.sep + outputfile_prefix + '_family_fraction_counts.txt' , 'w') for key in familyfractionalcounts.keys(): fout2.write(str(key) + '\t' + str(familyfractionalcounts[key])+ '\n') fout3 = open(outputfolder + os.path.sep + outputfile_prefix + '_fraction_counts.txt' , 'w') for key in fractionalcounts.keys(): fout3.write(str(key) + '\t' + repeatclass[key] + '\t' + repeatfamily[key] + '\t' + str(int(fractionalcounts[key])) + '\n') fout1.close() fout2.close() fout3.close() ################################################################################ # Remove Large intermediate files if os.path.exists(outputfolder + os.path.sep + outputfile_prefix + '_regionsorter.txt'): os.remove(outputfolder + os.path.sep + outputfile_prefix + '_regionsorter.txt') if os.path.exists(outputfolder + os.path.sep + 'pair1_bowtie'): shutil.rmtree(outputfolder + os.path.sep + 'pair1_bowtie') if os.path.exists(outputfolder + os.path.sep + 'pair2_bowtie'): shutil.rmtree(outputfolder + os.path.sep + 'pair2_bowtie') if os.path.exists(outputfolder + os.path.sep + 'sorted_bowtie'): shutil.rmtree(outputfolder + os.path.sep + 'sorted_bowtie') print ("... Done")