Mercurial > repos > miller-lab > genome_diversity
view select_snps.py @ 0:2c498d40ecde
Uploaded
author | miller-lab |
---|---|
date | Mon, 09 Apr 2012 12:03:06 -0400 |
parents | |
children |
line wrap: on
line source
#!/usr/bin/env python import os import sys import math from optparse import OptionParser import genome_diversity as gd def main_function(parse_arguments=None): if parse_arguments is None: parse_arguments = lambda arguments: (None, arguments) def main_decorator(to_decorate): def decorated_main(arguments=None): if arguments is None: arguments = sys.argv options, arguments = parse_arguments(arguments) sys.exit(to_decorate(options, arguments)) return decorated_main return main_decorator def parse_arguments(arguments): parser = OptionParser() parser.add_option('--input', dest='input') parser.add_option('--output', dest='output') parser.add_option('--index_dir', dest='index_dir') parser.add_option('--num_snps', dest='num_snps') parser.add_option('--ref_chrom_col', dest='ref_chrom_col') parser.add_option('--ref_pos_col', dest='ref_pos_col') parser.add_option('--ref_species', dest='ref_species') return parser.parse_args(arguments[1:]) @main_function(parse_arguments) def main(options, arguments): ref_chrom_idx = to_int( options.ref_chrom_col ) -1 ref_pos_idx = to_int( options.ref_pos_col ) -1 if (ref_chrom_idx < 1) or (ref_pos_idx < 1) or (ref_chrom_idx == ref_pos_idx): print >> sys.stderr, "Cannot locate reference genome sequence (ref) or reference genome position (rPos) column for this dataset." sys.exit(1) chrom_len_root = os.path.join( options.index_dir, 'shared/ucsc/chrom') chrom_len_file = '%s.len' % options.ref_species chrom_len_path = os.path.join(chrom_len_root, chrom_len_file) chrlens = gd.ChrLens( chrom_len_path ) total_len = 0 for chrom in chrlens: total_len += chrlens.length(chrom) total_requested = int( options.num_snps ) lines, data, comments = get_snp_lines_data_and_comments( options.input, ref_chrom_idx, ref_pos_idx ) selected = select_snps( data, total_len, total_requested ) out_data = fix_selection_and_order_like_input(data, selected, total_requested) write_selected_snps( options.output, out_data, lines, comments ) def to_int( value ): try: int_value = int( value ) except ValueError: int_value = 0 return int_value def get_snp_lines_data_and_comments( filename, chrom_idx, pos_idx ): fh = open( filename, 'r' ) if (chrom_idx >= pos_idx): needed = chrom_idx + 1 else: needed = pos_idx + 1 lines = [] data = [] comments = [] line_idx = 0 line_num = 0 for line in fh: line_num += 1 line = line.rstrip('\r\n') if line: if line.startswith('#'): comments.append(line) else: elems = line.split('\t') if len(elems) >= needed: chrom = elems[chrom_idx] try: pos = int(elems[pos_idx]) except ValueError: sys.stderr.write( "bad reference position in line %d column %d: %s\n" % ( line_num, pos_idx+1, elems[pos_idx] ) ) sys.exit(1) lines.append(line) chrom_sort = chrom.lstrip('chr') data.append( [chrom_sort, chrom, pos, line_num, line_idx] ) line_idx += 1 fh.close() data = sorted( data, key=lambda x: (x[0], x[2]) ) return lines, data, comments def select_snps( data, total_len, requested ): old_chrom = None next_print = 0 selected = [] space = total_len / requested for data_idx, datum in enumerate( data ): chrom = datum[1] pos = datum[2] if chrom != old_chrom: old_chrom = chrom next_print = 0 if pos >= next_print: selected.append(data_idx) next_print += space return selected def fix_selection_and_order_like_input(data, selected, requested): total_selected = len( selected ) a = float( total_selected ) / requested b = a / 2 idx_list = [] for i in range( requested ): idx = int( math.ceil( i * a + b ) - 1 ) idx_list.append( idx ) out_data = [] for i, data_idx in enumerate(selected): if total_selected > requested: if i in idx_list: out_data.append(data[data_idx]) else: out_data.append(data[data_idx]) out_data = sorted( out_data, key=lambda x: x[3] ) return out_data def write_selected_snps( filename, data, lines, comments ): fh = open( filename, 'w' ) for comment in comments: fh.write("%s\n" % comment ) for datum in data: line_idx = datum[4] fh.write("%s\n" % lines[line_idx]) fh.close() if __name__ == "__main__": main()