Mercurial > repos > iuc > concoct_extract_fasta_bins
comparison cut_up_fasta.py @ 1:a04028a8181d draft
"planemo upload for repository https://github.com/galaxyproject/tools-iuc/tree/master/tools/concoct commit 40a09cbfd6052f7b0295946621db1bdf58228b09"
author | iuc |
---|---|
date | Sun, 13 Mar 2022 08:44:08 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
0:1bc8fd1b3ed0 | 1:a04028a8181d |
---|---|
1 #!/usr/bin/env python | |
2 | |
3 import argparse | |
4 import gzip | |
5 from functools import partial | |
6 | |
7 from Bio import SeqIO | |
8 | |
9 | |
10 def cut_up_fasta(input_fasta, chunk_size, overlap, merge_last, output_fasta, output_bed, gzipped): | |
11 if gzipped: | |
12 _open = partial(gzip.open, mode='rt') | |
13 else: | |
14 _open = open | |
15 | |
16 fasta_fh = open(output_fasta, 'w') | |
17 | |
18 if output_bed is not None: | |
19 bed_fh = open(output_bed, 'w') | |
20 | |
21 with _open(input_fasta) as input_fh: | |
22 for record in SeqIO.parse(input_fh, "fasta"): | |
23 if (not merge_last and len(record.seq) > chunk_size) or (merge_last and len(record.seq) >= 2 * chunk_size): | |
24 for index, split_seq in enumerate(chunks(record.seq, chunk_size, overlap, merge_last)): | |
25 fasta_fh.write(f">{record.id}.concoct_part_{index}\n{split_seq}\n") | |
26 if output_bed is not None: | |
27 bed_fh.write(f"{record.id}\t{chunk_size * index}\t{chunk_size * index + len(split_seq)}\t{record.id}.concoct_part_{index}\n") | |
28 else: | |
29 fasta_fh.write(f">{record.id}.concoct_part_0\n{record.seq}\n") | |
30 if output_bed is not None: | |
31 bed_fh.write(f"{record.id}\t0\t{len(record.seq)}\t{record.id}.concoct_part_0\n") | |
32 if output_bed is not None: | |
33 bed_fh.close() | |
34 | |
35 | |
36 def chunks(seq, chunk_size, overlap_size, merge_last): | |
37 # Yield successive chunk_size-sized chunks from seq | |
38 # with given overlap overlap_size between the chunks. | |
39 assert chunk_size > overlap_size | |
40 if merge_last: | |
41 for i in range(0, len(seq) - chunk_size + 1, chunk_size - overlap_size): | |
42 yield seq[i:i + chunk_size] if i + chunk_size + chunk_size - overlap_size <= len(seq) else seq[i:] | |
43 else: | |
44 for i in range(0, len(seq), chunk_size - overlap_size): | |
45 yield seq[i:i + chunk_size] | |
46 | |
47 | |
48 parser = argparse.ArgumentParser() | |
49 parser.add_argument("--input_fasta", action="store", dest="input_fasta", help="Fasta files with contigs") | |
50 parser.add_argument("--gzipped", action="store_true", dest="gzipped", help="Input file is gzipped") | |
51 parser.add_argument("--chunk_size", action="store", dest="chunk_size", type=int, help="Chunk size\n") | |
52 parser.add_argument("--overlap_size", action="store", dest="overlap_size", type=int, help="Overlap size\n") | |
53 parser.add_argument("--merge_last", default=False, action="store_true", dest="merge_last", help="Concatenate final part to last contig\n") | |
54 parser.add_argument("--output_bed", action="store", dest="output_bed", default=None, help="BED file to be created with exact regions of the original contigs corresponding to the newly created contigs") | |
55 parser.add_argument("--output_fasta", action="store", dest="output_fasta", help="Output fasta file with cut contigs") | |
56 | |
57 args = parser.parse_args() | |
58 cut_up_fasta(args.input_fasta, args.chunk_size, args.overlap_size, args.merge_last, args.output_fasta, args.output_bed, args.gzipped) |