Mercurial > repos > modencode-dcc > macs
view macs_wrapper.py @ 13:0939725966c7 draft
Deleted selected files
author | modencode-dcc |
---|---|
date | Fri, 18 Jan 2013 15:24:36 -0500 |
parents | 426a52a01655 |
children |
line wrap: on
line source
import sys, subprocess, tempfile, shutil, glob, os, os.path, gzip from galaxy import eggs import pkg_resources pkg_resources.require( "simplejson" ) import simplejson CHUNK_SIZE = 1024 def gunzip_cat_glob_path( glob_path, target_filename, delete = False ): out = open( target_filename, 'wb' ) for filename in glob.glob( glob_path ): fh = gzip.open( filename, 'rb' ) while True: data = fh.read( CHUNK_SIZE ) if data: out.write( data ) else: break fh.close() if delete: os.unlink( filename ) out.close() def xls_to_interval( xls_file, interval_file, header = None ): out = open( interval_file, 'wb' ) if header: out.write( '#%s\n' % header ) wrote_header = False #From macs readme: Coordinates in XLS is 1-based which is different with BED format. for line in open( xls_file ): #keep all existing comment lines if line.startswith( '#' ): out.write( line ) elif not wrote_header: out.write( '#%s' % line ) wrote_header = True else: fields = line.split( '\t' ) if len( fields ) > 1: fields[1] = str( int( fields[1] ) - 1 ) out.write( '\t'.join( fields ) ) out.close() def main(): options = simplejson.load( open( sys.argv[1] ) ) output_bed = sys.argv[2] output_extra_html = sys.argv[3] output_extra_path = sys.argv[4] # Added peaks output argument output_peaks = sys.argv[5] experiment_name = '_'.join( options['experiment_name'].split() ) #save experiment name here, it will be used by macs for filenames (gzip of wig files will fail with spaces - macs doesn't properly escape them)..need to replace all whitespace, split makes this easier cmdline = "macs -t %s" % ",".join( options['input_chipseq'] ) if options['input_control']: cmdline = "%s -c %s" % ( cmdline, ",".join( options['input_control'] ) ) cmdline = "%s --format='%s' --name='%s' --gsize='%s' --tsize='%s' --bw='%s' --pvalue='%s' --mfold='%s' %s --lambdaset='%s' %s" % ( cmdline, options['format'], experiment_name, options['gsize'], options['tsize'], options['bw'], options['pvalue'], options['mfold'], options['nolambda'], options['lambdaset'], options['futurefdr'] ) if 'wig' in options: wigextend = int( options['wig']['wigextend'] ) if wigextend >= 0: wigextend = "--wigextend='%s'" % wigextend else: wigextend = '' cmdline = "%s --wig %s --space='%s'" % ( cmdline, wigextend, options['wig']['space'] ) if 'nomodel' in options: cmdline = "%s --nomodel --shiftsize='%s'" % ( cmdline, options['nomodel'] ) if 'diag' in options: cmdline = "%s --diag --fe-min='%s' --fe-max='%s' --fe-step='%s'" % ( cmdline, options['diag']['fe-min'], options['diag']['fe-max'], options['diag']['fe-step'] ) tmp_dir = tempfile.mkdtemp() #macs makes very messy output, need to contain it into a temp dir, then provide to user stderr_name = tempfile.NamedTemporaryFile().name # redirect stderr here, macs provides lots of info via stderr, make it into a report proc = subprocess.Popen( args=cmdline, shell=True, cwd=tmp_dir, stderr=open( stderr_name, 'wb' ) ) proc.wait() #We don't want to set tool run to error state if only warnings or info, e.g. mfold could be decreased to improve model, but let user view macs log #Do not terminate if error code, allow dataset (e.g. log) creation and cleanup if proc.returncode: stderr_f = open( stderr_name ) while True: chunk = stderr_f.read( CHUNK_SIZE ) if not chunk: stderr_f.close() break sys.stderr.write( chunk ) #run R to create pdf from model script if os.path.exists( os.path.join( tmp_dir, "%s_model.r" % experiment_name ) ): cmdline = 'R --vanilla --slave < "%s_model.r" > "%s_model.r.log"' % ( experiment_name, experiment_name ) proc = subprocess.Popen( args=cmdline, shell=True, cwd=tmp_dir ) proc.wait() #move bed out to proper output file created_bed_name = os.path.join( tmp_dir, "%s_peaks.bed" % experiment_name ) if os.path.exists( created_bed_name ): shutil.move( created_bed_name, output_bed ) #OICR peak_xls file created_peak_xls_file = os.path.join( tmp_dir, "%s_peaks.xls" % experiment_name ) if os.path.exists( created_peak_xls_file ): # shutil.copy( created_peak_xls_file, os.path.join ( "/mnt/galaxyData/tmp/", "%s_peaks.xls" % ( os.path.basename(output_extra_path) ))) shutil.copyfile( created_peak_xls_file, output_peaks ) #parse xls files to interval files as needed if options['xls_to_interval']: create_peak_xls_file = os.path.join( tmp_dir, '%s_peaks.xls' % experiment_name ) if os.path.exists( create_peak_xls_file ): xls_to_interval( create_peak_xls_file, options['xls_to_interval']['peaks_file'], header = 'peaks file' ) create_peak_xls_file = os.path.join( tmp_dir, '%s_negative_peaks.xls' % experiment_name ) if os.path.exists( create_peak_xls_file ): xls_to_interval( create_peak_xls_file, options['xls_to_interval']['negative_peaks_file'], header = 'negative peaks file' ) #merge and move wig files as needed, delete gz'd files and remove emptied dirs if 'wig' in options: wig_base_dir = os.path.join( tmp_dir, "%s_MACS_wiggle" % experiment_name ) if os.path.exists( wig_base_dir ): #treatment treatment_dir = os.path.join( wig_base_dir, "treat" ) if os.path.exists( treatment_dir ): gunzip_cat_glob_path( os.path.join( treatment_dir, "*.wig.gz" ), options['wig']['output_treatment_file'], delete = True ) os.rmdir( treatment_dir ) #control if options['input_control']: control_dir = os.path.join( wig_base_dir, "control" ) if os.path.exists( control_dir ): gunzip_cat_glob_path( os.path.join( control_dir, "*.wig.gz" ), options['wig']['output_control_file'], delete = True ) os.rmdir( control_dir ) os.rmdir( wig_base_dir ) #move all remaining files to extra files path of html file output to allow user download out_html = open( output_extra_html, 'wb' ) out_html.write( '<html><head><title>Additional output created by MACS (%s)</title></head><body><h3>Additional Files:</h3><p><ul>\n' % experiment_name ) os.mkdir( output_extra_path ) for filename in sorted( os.listdir( tmp_dir ) ): shutil.move( os.path.join( tmp_dir, filename ), os.path.join( output_extra_path, filename ) ) out_html.write( '<li><a href="%s">%s</a></li>\n' % ( filename, filename ) ) #out_html.write( '<li><a href="%s">%s</a>peakxls %s SomethingDifferent tmp_dir %s path %s exp_name %s</li>\n' % ( created_peak_xls_file, filename, filename, tmp_dir, output_extra_path, experiment_name ) ) out_html.write( '</ul></p>\n' ) out_html.write( '<h3>Messages from MACS:</h3>\n<p><pre>%s</pre></p>\n' % open( stderr_name, 'rb' ).read() ) out_html.write( '</body></html>\n' ) out_html.close() os.unlink( stderr_name ) os.rmdir( tmp_dir ) if __name__ == "__main__": main()