Mercurial > repos > gregory-minevich > check_snpeff_candidates
view checkSnpEffCandidates.py @ 9:f20585ccbf05 draft default tip
Uploaded
author | gregory-minevich |
---|---|
date | Mon, 08 Oct 2012 16:14:46 -0400 |
parents | a3873bb68495 |
children |
line wrap: on
line source
#!/usr/bin/python import sys import optparse import csv import re def main(): parser = optparse.OptionParser() parser.add_option('-s', '--snpeff_file', dest = 'snpeff_file', action = 'store', type = 'string', default = None, help = "Path to the snpEff file") parser.add_option('-c', '--candidate_list', dest = 'candidate_list', action = 'store', type = 'string', default = None, help = "Two column tabular list of candidate Gene ID, Type") parser.add_option('-o', '--output', dest = 'output', action = 'store', type = 'string', default = None, help = "Output file name") (options, args) = parser.parse_args() snpeff_file = options.snpeff_file candidate_list = options.candidate_list candidates = parse_candidate_list(candidate_list = candidate_list) mark_snpeff_file(snpeff_file = snpeff_file, output = options.output, candidates = candidates) def skip_and_write_headers(writer = None, reader = None, i_file = None): # count headers comment = 0 while reader.next()[0].startswith('#'): comment = comment + 1 # skip and write headers i_file.seek(0) for i in range(0, comment): row = reader.next() writer.writerow(row) def parse_candidate_list(candidate_list = ""): i_file = open(candidate_list, 'rU') reader = csv.reader(i_file, delimiter = '\t',) candidates = {} for row in reader: gene_id = row[0] gene_type = row[1] candidates[gene_id] = gene_type i_file.close() return candidates def mark_snpeff_file(snpeff_file = "", output = "", candidates = None): i_file = open(snpeff_file, 'rU') reader = csv.reader(i_file, delimiter = '\t') o_file = open(output, 'wb') writer = csv.writer(o_file, delimiter = '\t') skip_and_write_headers(writer = writer, reader = reader, i_file = i_file) for row in reader: gene_id = row[9] if gene_id in candidates: gene_type = candidates[gene_id] row.append(gene_type) else: row.append('') writer.writerow(row) o_file.close() i_file.close() if __name__ == "__main__": main()