Mercurial > repos > bjoern-gruening > bismark
view bismark_genome_preparation @ 0:36d124f44c0a draft
inital commit
author | bjoern-gruening |
---|---|
date | Tue, 25 Dec 2012 05:45:46 -0500 |
parents | |
children |
line wrap: on
line source
#!/usr/bin/perl -- use strict; use warnings; use Cwd; use File::Path qw(rmtree); $|++; ## This program is Copyright (C) 2010-12, Felix Krueger (felix.krueger@bbsrc.ac.uk) ## This program is free software: you can redistribute it and/or modify ## it under the terms of the GNU General Public License as published by ## the Free Software Foundation, either version 3 of the License, or ## (at your option) any later version. ## This program is distributed in the hope that it will be useful, ## but WITHOUT ANY WARRANTY; without even the implied warranty of ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the ## GNU General Public License for more details. ## You should have received a copy of the GNU General Public License ## along with this program. If not, see <http://www.gnu.org/licenses/>. use Getopt::Long; use Cwd; my $verbose; my $help; my $version; my $man; my $path_to_bowtie; my $multi_fasta; my $single_fasta; my $bowtie2; my $bismark_version = 'v0.7.7'; GetOptions ('verbose' => \$verbose, 'help' => \$help, 'man' => \$man, 'version' => \$version, 'path_to_bowtie:s' => \$path_to_bowtie, 'single_fasta' => \$single_fasta, 'bowtie2' => \$bowtie2, ); my $genome_folder = shift @ARGV; # mandatory my $CT_dir; my $GA_dir; if ($help or $man){ print_helpfile(); exit; } if ($version){ print << "VERSION"; Bismark - Bisulfite Mapper and Methylation Caller. Bismark Genome Preparation Version: $bismark_version Copyright 2010-12 Felix Krueger, Babraham Bioinformatics www.bioinformatics.babraham.ac.uk/projects/ VERSION exit; } if ($single_fasta){ print "Writing individual genomes out into single-entry fasta files (one per chromosome)\n\n"; $multi_fasta = 0; } else{ print "Writing bisulfite genomes out into a single MFA (multi FastA) file\n\n"; $single_fasta = 0; $multi_fasta = 1; } my @filenames = create_bisulfite_genome_folders(); process_sequence_files (); launch_bowtie_indexer(); sub launch_bowtie_indexer{ if ($bowtie2){ print "Bismark Genome Preparation - Step III: Launching the Bowtie 2 indexer\n"; } else{ print "Bismark Genome Preparation - Step III: Launching the Bowtie (1) indexer\n"; } print "Please be aware that this process can - depending on genome size - take up to several hours!\n"; sleep(5); ### if the path to bowtie was specfified explicitely if ($path_to_bowtie){ if ($bowtie2){ $path_to_bowtie =~ s/$/bowtie2-build/; } else{ $path_to_bowtie =~ s/$/bowtie-build/; } } ### otherwise we assume that bowtie-build is in the path else{ if ($bowtie2){ $path_to_bowtie = 'bowtie2-build'; } else{ $path_to_bowtie = 'bowtie-build'; } } $verbose and print "\n"; ### Forking the program to run 2 instances of Bowtie-build or Bowtie2-build (= the Bowtie (1/2) indexer) my $pid = fork(); # parent process if ($pid){ sleep(1); chdir $CT_dir or die "Unable to change directory: $!\n"; $verbose and warn "Preparing indexing of CT converted genome in $CT_dir\n"; my @fasta_files = <*.fa>; my $file_list = join (',',@fasta_files); $verbose and print "Parent process: Starting to index C->T converted genome with the following command:\n\n"; $verbose and print "$path_to_bowtie -f $file_list BS_CT\n\n"; sleep (11); exec ("$path_to_bowtie","-f","$file_list","BS_CT"); } # child process elsif ($pid == 0){ sleep(2); chdir $GA_dir or die "Unable to change directory: $!\n"; $verbose and warn "Preparing indexing of GA converted genome in $GA_dir\n"; my @fasta_files = <*.fa>; my $file_list = join (',',@fasta_files); $verbose and print "Child process: Starting to index G->A converted genome with the following command:\n\n"; $verbose and print "$path_to_bowtie -f $file_list BS_GA\n\n"; $verbose and print "(starting in 10 seconds)\n"; sleep(10); exec ("$path_to_bowtie","-f","$file_list","BS_GA"); } # if the platform doesn't support the fork command we will run the indexing processes one after the other else{ print "Forking process was not successful, therefore performing the indexing sequentially instead\n"; sleep(10); ### moving to CT genome folder $verbose and warn "Preparing to index CT converted genome in $CT_dir\n"; chdir $CT_dir or die "Unable to change directory: $!\n"; my @fasta_files = <*.fa>; my $file_list = join (',',@fasta_files); $verbose and print "$file_list\n\n"; sleep(2); system ("$path_to_bowtie","-f","$file_list","BS_CT"); @fasta_files=(); $file_list= ''; ### moving to GA genome folder $verbose and warn "Preparing to index GA converted genome in $GA_dir\n"; chdir $GA_dir or die "Unable to change directory: $!\n"; @fasta_files = <*.fa>; $file_list = join (',',@fasta_files); $verbose and print "$file_list\n\n"; sleep(2); exec ("$path_to_bowtie","-f","$file_list","BS_GA"); } } sub process_sequence_files { my ($total_CT_conversions,$total_GA_conversions) = (0,0); $verbose and print "Bismark Genome Preparation - Step II: Bisulfite converting reference genome\n\n"; sleep (3); $verbose and print "conversions performed:\n"; $verbose and print join("\t",'chromosome','C->T','G->A'),"\n"; ### If someone wants to index a genome which consists of thousands of contig and scaffold files we need to write the genome conversions into an MFA file ### Otherwise the list of comma separated chromosomes we provide for bowtie-build will get too long for the kernel to handle ### This is now the default option if ($multi_fasta){ ### Here we just use one multi FastA file name, append .CT_conversion or .GA_conversion and print all sequence conversions into these files my $bisulfite_CT_conversion_filename = "$CT_dir/genome_mfa.CT_conversion.fa"; open (CT_CONVERT,'>',$bisulfite_CT_conversion_filename) or die "Can't write to file $bisulfite_CT_conversion_filename: $!\n"; my $bisulfite_GA_conversion_filename = "$GA_dir/genome_mfa.GA_conversion.fa"; open (GA_CONVERT,'>',$bisulfite_GA_conversion_filename) or die "Can't write to file $bisulfite_GA_conversion_filename: $!\n"; } foreach my $filename(@filenames){ my ($chromosome_CT_conversions,$chromosome_GA_conversions) = (0,0); open (IN,$filename) or die "Failed to read from sequence file $filename $!\n"; # warn "Reading chromosome information from $filename\n\n"; ### first line needs to be a fastA header my $first_line = <IN>; chomp $first_line; ### Extracting chromosome name from the FastA header my $chromosome_name = extract_chromosome_name($first_line); ### alternatively, chromosomes can be written out into single-entry FastA files. This will only work for genomes with up to a few hundred chromosomes. unless ($multi_fasta){ my $bisulfite_CT_conversion_filename = "$CT_dir/$chromosome_name"; $bisulfite_CT_conversion_filename =~ s/$/.CT_conversion.fa/; open (CT_CONVERT,'>',$bisulfite_CT_conversion_filename) or die "Can't write to file $bisulfite_CT_conversion_filename: $!\n"; my $bisulfite_GA_conversion_filename = "$GA_dir/$chromosome_name"; $bisulfite_GA_conversion_filename =~ s/$/.GA_conversion.fa/; open (GA_CONVERT,'>',$bisulfite_GA_conversion_filename) or die "Can't write to file $bisulfite_GA_conversion_filename: $!\n"; } print CT_CONVERT ">",$chromosome_name,"_CT_converted\n"; # first entry print GA_CONVERT ">",$chromosome_name,"_GA_converted\n"; # first entry while (<IN>){ ### in case the line is a new fastA header if ($_ =~ /^>/){ ### printing out the stats for the previous chromosome $verbose and print join ("\t",$chromosome_name,$chromosome_CT_conversions,$chromosome_GA_conversions),"\n"; ### resetting the chromosome transliteration counters ($chromosome_CT_conversions,$chromosome_GA_conversions) = (0,0); ### Extracting chromosome name from the additional FastA header $chromosome_name = extract_chromosome_name($_); ### alternatively, chromosomes can be written out into single-entry FastA files. This will only work for genomes with up to a few hundred chromosomes. unless ($multi_fasta){ my $bisulfite_CT_conversion_filename = "$CT_dir/$chromosome_name"; $bisulfite_CT_conversion_filename =~ s/$/.CT_conversion.fa/; open (CT_CONVERT,'>',$bisulfite_CT_conversion_filename) or die "Can't write to file $bisulfite_CT_conversion_filename: $!\n"; my $bisulfite_GA_conversion_filename = "$GA_dir/$chromosome_name"; $bisulfite_GA_conversion_filename =~ s/$/.GA_conversion.fa/; open (GA_CONVERT,'>',$bisulfite_GA_conversion_filename) or die "Can't write to file $bisulfite_GA_conversion_filename: $!\n"; } print CT_CONVERT ">",$chromosome_name,"_CT_converted\n"; print GA_CONVERT ">",$chromosome_name,"_GA_converted\n"; } else{ my $sequence = uc$_; ### (I) First replacing all ambiguous sequence characters (such as M,S,R....) by N (G,A,T,C,N and the line endings \r and \n are added to a character group) $sequence =~ s/[^ATCGN\n\r]/N/g; ### (II) Writing the chromosome out into a C->T converted version (equals forward strand conversion) my $CT_sequence = $sequence; my $CT_transliterations_performed = ($CT_sequence =~ tr/C/T/); # converts all Cs into Ts $total_CT_conversions += $CT_transliterations_performed; $chromosome_CT_conversions += $CT_transliterations_performed; print CT_CONVERT $CT_sequence; ### (III) Writing the chromosome out in a G->A converted version of the forward strand (this is equivalent to reverse- ### complementing the forward strand and then C->T converting it) my $GA_sequence = $sequence; my $GA_transliterations_performed = ($GA_sequence =~ tr/G/A/); # converts all Gs to As on the forward strand $total_GA_conversions += $GA_transliterations_performed; $chromosome_GA_conversions += $GA_transliterations_performed; print GA_CONVERT $GA_sequence; } } $verbose and print join ("\t",$chromosome_name,$chromosome_CT_conversions,$chromosome_GA_conversions),"\n"; } close (CT_CONVERT) or die "Failed to close filehandle: $!\n"; close (GA_CONVERT) or die "Failed to close filehandle: $!\n"; print "\nTotal number of conversions performed:\n"; print "C->T:\t$total_CT_conversions\n"; print "G->A:\t$total_GA_conversions\n"; warn "\nStep II - Genome bisulfite conversions - completed\n\n\n"; } sub extract_chromosome_name { my $header = shift; ## Bowtie extracts the first string after the initial > in the FASTA file, so we are doing this as well if ($header =~ s/^>//){ my ($chromosome_name) = split (/\s+/,$header); return $chromosome_name; } else{ die "The specified chromosome file doesn't seem to be in FASTA format as required! $!\n"; } } sub create_bisulfite_genome_folders{ $verbose and print "Bismark Genome Preparation - Step I: Preparing folders\n\n"; # Ensuring a genome folder has been specified if ($genome_folder){ unless ($genome_folder =~ /\/$/){ $genome_folder =~ s/$/\//; } $verbose and print "Path to genome folder specified: $genome_folder\n"; chdir $genome_folder or die "Could't move to directory $genome_folder. Make sure the directory exists! $!"; # making the genome folder path abolsolute so it won't break if the path was specified relative $genome_folder = getcwd; unless ($genome_folder =~ /\/$/){ $genome_folder =~ s/$/\//; } } else{ $verbose and print "Genome folder was not provided as argument "; while (1){ print "Please specify a genome folder to be bisulfite converted:\n"; $genome_folder = <STDIN>; chomp $genome_folder; # adding a trailing slash unless already present unless ($genome_folder =~ /\/$/){ $genome_folder =~ s/$/\//; } if (chdir $genome_folder){ last; } else{ warn "Could't move to directory $genome_folder! $!"; } } } if ($path_to_bowtie){ unless ($path_to_bowtie =~ /\/$/){ $path_to_bowtie =~ s/$/\//; } if (chdir $path_to_bowtie){ if ($bowtie2){ $verbose and print "Path to Bowtie 2 specified: $path_to_bowtie\n"; } else{ $verbose and print "Path to Bowtie (1) specified: $path_to_bowtie\n"; } } else{ die "There was an error with the path to bowtie: $!\n"; } } chdir $genome_folder or die "Could't move to directory $genome_folder. Make sure the directory exists! $!"; # Exiting unless there are fastA files in the folder my @filenames = <*.fa>; ### if there aren't any genomic files with the extension .fa we will look for files with the extension .fasta unless (@filenames){ @filenames = <*.fasta>; } unless (@filenames){ die "The specified genome folder $genome_folder does not contain any sequence files in FastA format (with .fa or .fasta file extensions\n"; } warn "Bisulfite Genome Indexer version $bismark_version (last modified 17 Nov 2011)\n\n"; sleep (3); # creating a directory inside the genome folder to store the bisfulfite genomes unless it already exists my $bisulfite_dir = "${genome_folder}Bisulfite_Genome/"; unless (-d $bisulfite_dir){ mkdir $bisulfite_dir or die "Unable to create directory $bisulfite_dir $!\n"; $verbose and print "Created Bisulfite Genome folder $bisulfite_dir\n"; } else{ while (1){ print "\nA directory called $bisulfite_dir already exists. Bisulfite converted sequences and/or already existing Bowtie (1 or 2) indexes might be overwritten!\nDo you want to continue anyway?\t"; my $proceed = <STDIN>; chomp $proceed; if ($proceed =~ /^y/i ){ last; } elsif ($proceed =~ /^n/i){ die "Terminated by user\n\n"; } } } ### as of version 0.6.0 the Bismark indexer will no longer delete the Bisulfite_Genome directory if it was present already, since it could store the Bowtie 1 or 2 indexes already # removing any existing files and subfolders in the bisulfite directory (the specified directory won't be deleted) # rmtree($bisulfite_dir, {verbose => 1,keep_root => 1}); # unless (-d $bisulfite_dir){ # had to add this after changing remove_tree to rmtree // suggested by Samantha Cooper @ Illumina # mkdir $bisulfite_dir or die "Unable to create directory $bisulfite_dir $!\n"; # } # } chdir $bisulfite_dir or die "Unable to move to $bisulfite_dir\n"; $CT_dir = "${bisulfite_dir}CT_conversion/"; $GA_dir = "${bisulfite_dir}GA_conversion/"; # creating 2 subdirectories to store a C->T (forward strand conversion) and a G->A (reverse strand conversion) # converted version of the genome unless (-d $CT_dir){ mkdir $CT_dir or die "Unable to create directory $CT_dir $!\n"; $verbose and print "Created Bisulfite Genome folder $CT_dir\n"; } unless (-d $GA_dir){ mkdir $GA_dir or die "Unable to create directory $GA_dir $!\n"; $verbose and print "Created Bisulfite Genome folder $GA_dir\n"; } # moving back to the original genome folder chdir $genome_folder or die "Could't move to directory $genome_folder $!"; # $verbose and print "Moved back to genome folder folder $genome_folder\n"; warn "\nStep I - Prepare genome folders - completed\n\n\n"; return @filenames; } sub print_helpfile{ print << 'HOW_TO'; DESCRIPTION This script is supposed to convert a specified reference genome into two different bisulfite converted versions and index them for alignments with Bowtie 1 (default), or Bowtie 2. The first bisulfite genome will have all Cs converted to Ts (C->T), and the other one will have all Gs converted to As (G->A). Both bisulfite genomes will be stored in subfolders within the reference genome folder. Once the bisulfite conversion has been completed the program will fork and launch two simultaneous instances of the bowtie 1 or 2 indexer (bowtie-build or bowtie2-build). Be aware that the indexing process can take up to several hours; this will mainly depend on genome size and system resources. The following is a brief description of command line options and arguments to control the Bismark Genome Preparation script: USAGE: bismark_genome_preparation [options] <arguments> OPTIONS: --help/--man Displays this help filea and exits. --version Displays version information and exits. --verbose Print verbose output for more details or debugging. --path_to_bowtie The full path to the Bowtie 1 or Bowtie 2 installation on your system.If the path </../../> is not provided as an option you will be prompted for it. --bowtie2 This will create bisulfite indexes for Bowtie 2. (Default: Bowtie 1). --single_fasta Instruct the Bismark Indexer to write the converted genomes into single-entry FastA files instead of making one multi-FastA file (MFA) per chromosome. This might be useful if individual bisulfite converted chromosomes are needed (e.g. for debugging), however it can cause a problem with indexing if the number of chromosomes is vast (this is likely to be in the range of several thousand files; the operating system can only handle lists up to a certain length, and some newly assembled genomes may contain 20000-50000 contigs of scaffold files which do exceed this list length limit). ARGUMENTS: <path_to_genome_folder> The path to the folder containing the genome to be bisulfite converted. At the current time Bismark Genome Preparation expects one or more fastA files in the folder (with the file extension: .fa or .fasta). If the path is not provided as an argument you will be prompted for it. This script was last modified on 18 Nov 2011. HOW_TO }