Mercurial > repos > iuc > vsnp_determine_ref_from_data
view vsnp_determine_ref_from_data.py @ 5:a8560decb495 draft
"planemo upload for repository https://github.com/galaxyproject/tools-iuc/tree/master/tools/vsnp commit 2a94c64d6c7236550bf483d2ffc4e86248c63aab"
author | iuc |
---|---|
date | Tue, 16 Nov 2021 20:12:56 +0000 |
parents | 9b4011d5e8cb |
children |
line wrap: on
line source
#!/usr/bin/env python import argparse import gzip import os from collections import OrderedDict import yaml from Bio.SeqIO.QualityIO import FastqGeneralIterator OUTPUT_DBKEY_DIR = 'output_dbkey' OUTPUT_METRICS_DIR = 'output_metrics' def get_sample_name(file_path): base_file_name = os.path.basename(file_path) if base_file_name.find(".") > 0: # Eliminate the extension. return os.path.splitext(base_file_name)[0] return base_file_name def get_dbkey(dnaprints_dict, key, s): # dnaprints_dict looks something like this: # {'brucella': {'NC_002945v4': ['11001110', '11011110', '11001100']} # {'bovis': {'NC_006895': ['11111110', '00010010', '01111011']}} d = dnaprints_dict.get(key, {}) for data_table_value, v_list in d.items(): if s in v_list: return data_table_value return "" def get_dnaprints_dict(dnaprint_fields): # A dndprint_fields entry looks something liek this. # [['AF2122', '/galaxy/tool-data/vsnp/AF2122/dnaprints/NC_002945v4.yml']] dnaprints_dict = {} for item in dnaprint_fields: # Here item is a 2-element list of data # table components, # value and path. value = item[0] path = item[1].strip() with open(path, "rt") as fh: # The format of all dnaprints yaml # files is something like this: # brucella: # - 0111111111111111 print_dict = yaml.load(fh, Loader=yaml.Loader) for print_dict_k, print_dict_v in print_dict.items(): dnaprints_v_dict = dnaprints_dict.get(print_dict_k, {}) if len(dnaprints_v_dict) > 0: # dnaprints_dict already contains k (e.g., 'brucella', # and dnaprints_v_dict will be a dictionary # that # looks something like this: # {'NC_002945v4': ['11001110', '11011110', '11001100']} value_list = dnaprints_v_dict.get(value, []) value_list = value_list + print_dict_v dnaprints_v_dict[value] = value_list else: # dnaprints_v_dict is an empty dictionary. dnaprints_v_dict[value] = print_dict_v dnaprints_dict[print_dict_k] = dnaprints_v_dict # dnaprints_dict looks something like this: # {'brucella': {'NC_002945v4': ['11001110', '11011110', '11001100']} # {'bovis': {'NC_006895': ['11111110', '00010010', '01111011']}} return dnaprints_dict def get_group_and_dbkey(dnaprints_dict, brucella_string, brucella_sum, bovis_string, bovis_sum, para_string, para_sum): if brucella_sum > 3: group = "Brucella" dbkey = get_dbkey(dnaprints_dict, "brucella", brucella_string) elif bovis_sum > 3: group = "TB" dbkey = get_dbkey(dnaprints_dict, "bovis", bovis_string) elif para_sum >= 1: group = "paraTB" dbkey = get_dbkey(dnaprints_dict, "para", para_string) else: group = "" dbkey = "" return group, dbkey def get_oligo_dict(): oligo_dict = {} oligo_dict["01_ab1"] = "AATTGTCGGATAGCCTGGCGATAACGACGC" oligo_dict["02_ab3"] = "CACACGCGGGCCGGAACTGCCGCAAATGAC" oligo_dict["03_ab5"] = "GCTGAAGCGGCAGACCGGCAGAACGAATAT" oligo_dict["04_mel"] = "TGTCGCGCGTCAAGCGGCGTGAAATCTCTG" oligo_dict["05_suis1"] = "TGCGTTGCCGTGAAGCTTAATTCGGCTGAT" oligo_dict["06_suis2"] = "GGCAATCATGCGCAGGGCTTTGCATTCGTC" oligo_dict["07_suis3"] = "CAAGGCAGATGCACATAATCCGGCGACCCG" oligo_dict["08_ceti1"] = "GTGAATATAGGGTGAATTGATCTTCAGCCG" oligo_dict["09_ceti2"] = "TTACAAGCAGGCCTATGAGCGCGGCGTGAA" oligo_dict["10_canis4"] = "CTGCTACATAAAGCACCCGGCGACCGAGTT" oligo_dict["11_canis"] = "ATCGTTTTGCGGCATATCGCTGACCACAGC" oligo_dict["12_ovis"] = "CACTCAATCTTCTCTACGGGCGTGGTATCC" oligo_dict["13_ether2"] = "CGAAATCGTGGTGAAGGACGGGACCGAACC" oligo_dict["14_63B1"] = "CCTGTTTAAAAGAATCGTCGGAACCGCTCT" oligo_dict["15_16M0"] = "TCCCGCCGCCATGCCGCCGAAAGTCGCCGT" oligo_dict["16_mel1b"] = "TCTGTCCAAACCCCGTGACCGAACAATAGA" oligo_dict["17_tb157"] = "CTCTTCGTATACCGTTCCGTCGTCACCATGGTCCT" oligo_dict["18_tb7"] = "TCACGCAGCCAACGATATTCGTGTACCGCGACGGT" oligo_dict["19_tbbov"] = "CTGGGCGACCCGGCCGACCTGCACACCGCGCATCA" oligo_dict["20_tb5"] = "CCGTGGTGGCGTATCGGGCCCCTGGATCGCGCCCT" oligo_dict["21_tb2"] = "ATGTCTGCGTAAAGAAGTTCCATGTCCGGGAAGTA" oligo_dict["22_tb3"] = "GAAGACCTTGATGCCGATCTGGGTGTCGATCTTGA" oligo_dict["23_tb4"] = "CGGTGTTGAAGGGTCCCCCGTTCCAGAAGCCGGTG" oligo_dict["24_tb6"] = "ACGGTGATTCGGGTGGTCGACACCGATGGTTCAGA" oligo_dict["25_para"] = "CCTTTCTTGAAGGGTGTTCG" oligo_dict["26_para_sheep"] = "CGTGGTGGCGACGGCGGCGGGCCTGTCTAT" oligo_dict["27_para_cattle"] = "TCTCCTCGGTCGGTGATTCGGGGGCGCGGT" return oligo_dict def get_seq_counts(value, fastq_list, gzipped): count = 0 for fastq_file in fastq_list: if gzipped: with gzip.open(fastq_file, 'rt') as fh: for title, seq, qual in FastqGeneralIterator(fh): count += seq.count(value) else: with open(fastq_file, 'r') as fh: for title, seq, qual in FastqGeneralIterator(fh): count += seq.count(value) return(value, count) def get_species_counts(fastq_list, gzipped): count_summary = {} oligo_dict = get_oligo_dict() for v1 in oligo_dict.values(): returned_value, count = get_seq_counts(v1, fastq_list, gzipped) for key, v2 in oligo_dict.items(): if returned_value == v2: count_summary.update({key: count}) count_list = [] for v in count_summary.values(): count_list.append(v) brucella_sum = sum(count_list[:16]) bovis_sum = sum(count_list[16:24]) para_sum = sum(count_list[24:]) return count_summary, count_list, brucella_sum, bovis_sum, para_sum def get_species_strings(count_summary): binary_dictionary = {} for k, v in count_summary.items(): if v > 1: binary_dictionary.update({k: 1}) else: binary_dictionary.update({k: 0}) binary_dictionary = OrderedDict(sorted(binary_dictionary.items())) binary_list = [] for v in binary_dictionary.values(): binary_list.append(v) brucella_binary = binary_list[:16] brucella_string = ''.join(str(e) for e in brucella_binary) bovis_binary = binary_list[16:24] bovis_string = ''.join(str(e) for e in bovis_binary) para_binary = binary_list[24:] para_string = ''.join(str(e) for e in para_binary) return brucella_string, bovis_string, para_string def output_dbkey(file_name, dbkey, output_file): # Output the dbkey. with open(output_file, "w") as fh: fh.write("%s" % dbkey) def output_files(fastq_file, count_list, group, dbkey, dbkey_file, metrics_file): base_file_name = get_sample_name(fastq_file) output_dbkey(base_file_name, dbkey, dbkey_file) output_metrics(base_file_name, count_list, group, dbkey, metrics_file) def output_metrics(file_name, count_list, group, dbkey, output_file): # Output the metrics. with open(output_file, "w") as fh: fh.write("Sample: %s\n" % file_name) fh.write("Brucella counts: ") for i in count_list[:16]: fh.write("%d," % i) fh.write("\nTB counts: ") for i in count_list[16:24]: fh.write("%d," % i) fh.write("\nPara counts: ") for i in count_list[24:]: fh.write("%d," % i) fh.write("\nGroup: %s" % group) fh.write("\ndbkey: %s\n" % dbkey) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--dnaprint_fields', action='append', dest='dnaprint_fields', nargs=2, help="List of dnaprints data table value, name and path fields") parser.add_argument('--read1', action='store', dest='read1', help='Required: single read') parser.add_argument('--read2', action='store', dest='read2', required=False, default=None, help='Optional: paired read') parser.add_argument('--gzipped', action='store_true', dest='gzipped', help='Input files are gzipped') parser.add_argument('--output_dbkey', action='store', dest='output_dbkey', help='Output reference file') parser.add_argument('--output_metrics', action='store', dest='output_metrics', help='Output metrics file') args = parser.parse_args() fastq_list = [args.read1] if args.read2 is not None: fastq_list.append(args.read2) # The value of dnaprint_fields is a list of lists, where each list is # the [value, name, path] components of the vsnp_dnaprints data table. # The data_manager_vsnp_dnaprints tool assigns the dbkey column from the # all_fasta data table to the value column in the vsnp_dnaprints data # table to ensure a proper mapping for discovering the dbkey. dnaprints_dict = get_dnaprints_dict(args.dnaprint_fields) # Here fastq_list consists of either a single read # or a set of paired reads, producing single outputs. count_summary, count_list, brucella_sum, bovis_sum, para_sum = get_species_counts(fastq_list, args.gzipped) brucella_string, bovis_string, para_string = get_species_strings(count_summary) group, dbkey = get_group_and_dbkey(dnaprints_dict, brucella_string, brucella_sum, bovis_string, bovis_sum, para_string, para_sum) output_files(args.read1, count_list, group, dbkey, dbkey_file=args.output_dbkey, metrics_file=args.output_metrics)