Mercurial > repos > devteam > fastq_stats
view fastq_stats.py @ 0:9b7b4e0ca9db draft
Imported from capsule None
author | devteam |
---|---|
date | Mon, 27 Jan 2014 09:29:03 -0500 |
parents | |
children |
line wrap: on
line source
#Dan Blankenberg import sys from galaxy_utils.sequence.fastq import fastqReader, fastqAggregator VALID_NUCLEOTIDES = [ 'A', 'C', 'G', 'T', 'N' ] VALID_COLOR_SPACE = map( str, range( 7 ) ) + [ '.' ] SUMMARY_STAT_ORDER = ['read_count', 'min_score', 'max_score', 'sum_score', 'mean_score', 'q1', 'med_score', 'q3', 'iqr', 'left_whisker', 'right_whisker' ] def main(): input_filename = sys.argv[1] output_filename = sys.argv[2] input_type = sys.argv[3] or 'sanger' aggregator = fastqAggregator() num_reads = None fastq_read = None for num_reads, fastq_read in enumerate( fastqReader( open( input_filename ), format = input_type ) ): aggregator.consume_read( fastq_read ) out = open( output_filename, 'wb' ) valid_nucleotides = VALID_NUCLEOTIDES if fastq_read: if fastq_read.sequence_space == 'base': out.write( '#column\tcount\tmin\tmax\tsum\tmean\tQ1\tmed\tQ3\tIQR\tlW\trW\toutliers\tA_Count\tC_Count\tG_Count\tT_Count\tN_Count\tother_bases\tother_base_count\n' ) else: out.write( '#column\tcount\tmin\tmax\tsum\tmean\tQ1\tmed\tQ3\tIQR\tlW\trW\toutliers\t0_Count\t1_Count\t2_Count\t3_Count\t4_Count\t5_Count\t6_Count\t._Count\tother_bases\tother_base_count\n' ) valid_nucleotides = VALID_COLOR_SPACE for i in range( aggregator.get_max_read_length() ): column_stats = aggregator.get_summary_statistics_for_column( i ) out.write( '%i\t' % ( i + 1 ) ) out.write( '%s\t' * len( SUMMARY_STAT_ORDER ) % tuple( [ column_stats[ key ] for key in SUMMARY_STAT_ORDER ] ) ) out.write( '%s\t' % ','.join( map( str, column_stats['outliers'] ) ) ) base_counts = aggregator.get_base_counts_for_column( i ) for nuc in valid_nucleotides: out.write( "%s\t" % base_counts.get( nuc, 0 ) ) extra_nucs = sorted( [ nuc for nuc in base_counts.keys() if nuc not in valid_nucleotides ] ) out.write( "%s\t%s\n" % ( ','.join( extra_nucs ), ','.join( str( base_counts[nuc] ) for nuc in extra_nucs ) ) ) out.close() if num_reads is None: print "No valid fastq reads could be processed." else: print "%i fastq reads were processed." % ( num_reads + 1 ) print "Based upon quality values and sequence characters, the input data is valid for: %s" % ( ", ".join( aggregator.get_valid_formats() ) or "None" ) ascii_range = aggregator.get_ascii_range() decimal_range = aggregator.get_decimal_range() print "Input ASCII range: %s(%i) - %s(%i)" % ( repr( ascii_range[0] ), ord( ascii_range[0] ), repr( ascii_range[1] ), ord( ascii_range[1] ) ) #print using repr, since \x00 (null) causes info truncation in galaxy when printed print "Input decimal range: %i - %i" % ( decimal_range[0], decimal_range[1] ) if __name__ == "__main__": main()