EukPhylo/PTL2/Scripts/preguidance.py
2024-01-12 13:25:53 -05:00

115 lines
4.9 KiB
Python

import os, sys, re
from Bio import SeqIO
def run(params):
try:
ogs = list(dict.fromkeys([line.strip() for line in open(params.gf_list)]))
except (FileNotFoundError, TypeError) as e:
print('\nERROR: Unable to read GF list file. Please make sure that the path is correct and that the file is formatted correctly.\n\n' + str(e) + '\n')
exit()
try:
taxa = list(dict.fromkeys([line.strip() for line in open(params.taxon_list)]))
except (FileNotFoundError, TypeError) as e:
print('\nERROR: Unable to read taxon list file. Please make sure that the path is correct and that the file is formatted correctly.\n\n' + str(e) + '\n')
exit()
if params.sim_taxa != None:
try:
sim_taxa = list(dict.fromkeys([line.strip() for line in open(params.sim_taxa)]))
except (FileNotFoundError, TypeError) as e:
print('\nERROR: Unable to read similarity taxa list file. Please make sure that the path is correct and that the file is formatted correctly.\n\n' + str(e) + '\n')
exit()
else:
sim_taxa = 'all'
if params.blacklist != None:
try:
blacklist_seqs = list(dict.fromkeys([line.strip() for line in open(params.blacklist)]))
except (FileNotFoundError, TypeError) as e:
print('\nERROR: Unable to read blacklist file. Please make sure that the path is correct and that the file is formatted correctly.\n\n' + str(e) + '\n')
exit()
else:
blacklist_seqs = []
if not os.path.isdir(params.data):
print('\nInput amino-acid data files not found. Please make sure that the given path (--data) is correct.\n')
aa_files = [f for f in os.listdir(params.data) if f[:10] in taxa if f.endswith('.faa') or f.endswith('.fa') or f.endswith('.fasta')]
missing_taxa = [tax for tax in taxa if tax not in [f[:10] for f in aa_files]]
if(len(missing_taxa) > 0):
print('\nWARNING: The following taxa in the taxon list are missing amino-acid files in ' + params.data + ':\n' + '\n'.join(['\t' + t for t in missing_taxa]) + '\n')
os.mkdir(params.output + '/Output/Intermediate/SF_Diamond')
removed_file = open(params.output + '/Output/Pre-Guidance/SimFilter_removed.txt', 'w')
for og in ogs:
print('\nProcessing ' + og + '\n')
with open(params.output + '/Output/Pre-Guidance/' + og + '_preguidance.fasta', 'w') as preguidance_file:
for taxon_file in aa_files:
recs = []
for rec in sorted([rec for rec in SeqIO.parse(params.data + '/' + taxon_file, 'fasta') if rec.id[-10:] == og and rec.id not in blacklist_seqs and rec.id[-10:].startswith(params.og_identifier)], key=lambda x: -len(x.seq)):
if(rec.id == rec.description):
recs.append(rec)
else:
print('\n\tThe sequence ID ' + rec.description + ' is invalid. Please make sure that sequence IDs contain no spaces, tabs, etc. This sequence is being excluded.\n')
if sim_taxa == 'all':
use_taxon = True
else:
if taxon_file[:10] in sim_taxa:
use_taxon = True
else:
use_taxon = False
masters = []; removed = 0; flag = 0; cycle = 0
if params.similarity_filter and use_taxon:
if len(recs) > 1:
while flag == 0:
master_file_name = params.output + '/Output/Intermediate/SF_Diamond/' + og + '_' + taxon_file[:10] + '_master_' + str(cycle)
query_file_name = params.output + '/Output/Intermediate/SF_Diamond/' + og + '_' + taxon_file[:10] + '_queries_' + str(cycle) + '.fasta'
diamond_out_name = params.output + '/Output/Intermediate/SF_Diamond/' + og + '_' + taxon_file[:10] + '_diamond_results_' + str(cycle) + '.tsv'
open(master_file_name + '.faa', 'w').write('>' + recs[0].id + '\n' + str(recs[0].seq) + '\n\n')
masters.append(recs[0])
with open(query_file_name, 'w') as queries:
for rec in recs[1:]:
queries.write('>' + rec.id + '\n' + str(rec.seq) + '\n\n')
os.system('diamond makedb --in ' + master_file_name + '.faa -d ' + master_file_name)
os.system('diamond blastp -d ' + master_file_name + '.dmnd -q ' + query_file_name + ' --outfmt 6 -o ' + diamond_out_name)
diamond_out = open(diamond_out_name).readlines()
recs_to_remove = []
for line in diamond_out:
line = line.strip().split('\t')
if float(line[2])/100 >= params.sim_cutoff:
recs_to_remove.append(line[0]); removed =+ 1
if len([rec for rec in recs[1:] if rec.id not in recs_to_remove]) < 2:
recs = [rec for rec in recs[1:] if rec.id not in recs_to_remove]
flag = 1
else:
recs = [rec for rec in recs[1:] if rec.id not in recs_to_remove]
cycle += 1
for item in recs_to_remove:
removed_file.write(f"{item}\n")
print('\n\t' + str(removed) + ' sequence(s) removed by the similarity filter (' + str(cycle + 1) + ' iterations) from ' + taxon_file[:10] + '\n')
for rec in recs + masters:
preguidance_file.write('>' + rec.id + '\n' + str(rec.seq) + '\n\n')
removed_file.close()
if(not params.keep_temp):
os.system('rm -r ' + params.output + '/Output/Intermediate/SF_Diamond')