view generate-putative-isp.py @ 9:9fa911974de6 draft default tip

planemo upload commit a4912b5d710e0082b764a8c90bf184078e74d83d-dirty
author cpt
date Fri, 20 Sep 2024 03:38:42 +0000
parents bc16792ecfd2
children
line wrap: on
line source

#!/usr/bin/env python

##### findSpanin.pl --> findSpanin.py
######### Incooperated from the findSpanin.pl script, but better and more snakey.

import argparse
from cpt import OrfFinder
import regex as re
from spaninFuncs import *


if __name__ == "__main__":

    parser = argparse.ArgumentParser(
        description="Get putative protein candidates for spanins"
    )
    parser.add_argument(
        "fasta_file", type=argparse.FileType("r"), help="Fasta file"
    )  # the "input" argument

    parser.add_argument(
        "-f",
        "--format",
        dest="seq_format",
        default="fasta",
        help="Sequence format (e.g. fasta, fastq, sff)",
    )  # optional formats for input, currently just going to do ntFASTA

    parser.add_argument(
        "--strand",
        dest="strand",
        choices=("both", "forward", "reverse"),
        default="both",
        help="select strand",
    )  # Selection of +, -, or both strands

    parser.add_argument(
        "--table", dest="table", default=11, help="NCBI Translation table", type=int
    )  # Uses "default" NCBI codon table. This should always (afaik) be what we want...

    parser.add_argument(
        "-t",
        "--ftype",
        dest="ftype",
        choices=("CDS", "ORF"),
        default="ORF",
        help="Find ORF or CDSs",
    )  # "functional type(?)" --> Finds ORF or CDS, for this we want just the ORF

    parser.add_argument(
        "-e",
        "--ends",
        dest="ends",
        choices=("open", "closed"),
        default="closed",
        help="Open or closed. Closed ensures start/stop codons are present",
    )  # includes the start and stop codon

    parser.add_argument(
        "-m",
        "--mode",
        dest="mode",
        choices=("all", "top", "one"),
        default="all",  # I think we want this to JUST be all...nearly always
        help="Output all ORFs/CDSs from sequence, all ORFs/CDSs with max length, or first with maximum length",
    )

    parser.add_argument(
        "--switch",
        dest="switch",
        default="all",
        help="switch between ALL putative osps, or a range. If not all, insert a range of two integers separated by a colon (:). Eg: 1234:4321",
    )
    # isp parameters
    parser.add_argument(
        "--isp_min_len",
        dest="isp_min_len",
        default=60,
        help="Minimum ORF length, measured in codons",
        type=int,
    )
    parser.add_argument(
        "--isp_on",
        dest="out_isp_nuc",
        type=argparse.FileType("w"),
        default="_out_isp.fna",
        help="Output nucleotide sequences, FASTA",
    )
    parser.add_argument(
        "--isp_op",
        dest="out_isp_prot",
        type=argparse.FileType("w"),
        default="_out_isp.fa",
        help="Output protein sequences, FASTA",
    )
    parser.add_argument(
        "--isp_ob",
        dest="out_isp_bed",
        type=argparse.FileType("w"),
        default="_out_isp.bed",
        help="Output BED file",
    )
    parser.add_argument(
        "--isp_og",
        dest="out_isp_gff3",
        type=argparse.FileType("w"),
        default="_out_isp.gff3",
        help="Output GFF3 file",
    )
    parser.add_argument(
        "--isp_min_dist",
        dest="isp_min_dist",
        default=10,
        help="Minimal distance to first AA of TMD, measured in AA",
        type=int,
    )
    parser.add_argument(
        "--isp_max_dist",
        dest="isp_max_dist",
        default=30,
        help="Maximum distance to first AA of TMD, measured in AA",
        type=int,
    )
    parser.add_argument(
        "--putative_isp",
        dest="putative_isp_fa",
        type=argparse.FileType("w"),
        default="_putative_isp.fa",
        help="Output of putative FASTA file",
    )
    parser.add_argument(
        "--min_tmd_size",
        dest="min_tmd_size",
        default=10,
        help="Minimal size of the TMD domain",
        type=int,
    )
    parser.add_argument(
        "--max_tmd_size",
        dest="max_tmd_size",
        default=20,
        help="Maximum size of the TMD domain",
        type=int,
    )
    parser.add_argument(
        "--summary_isp_txt",
        dest="summary_isp_txt",
        type=argparse.FileType("w"),
        default="_summary_isp.txt",
        help="Summary statistics on putative i-spanins",
    )
    parser.add_argument(
        "--putative_isp_gff",
        dest="putative_isp_gff",
        type=argparse.FileType("w"),
        default="_putative_isp.gff3",
        help="gff3 output for putative i-spanins",
    )

    parser.add_argument(
        "--max_isp",
        dest="max_isp",
        default=230,
        help="Maximum size of the ISP",
        type=int,
    )

    parser.add_argument("--isp_mode", action="store_true", default=True)

    parser.add_argument(
        "--peri_min",
        type=int,
        default=18,
        help="amount of residues after TMD is found min",
    )

    parser.add_argument(
        "--peri_max",
        type=int,
        default=206,
        help="amount of residues after TMD is found max",
    )
    # parser.add_argument('-v', action='version', version='0.3.0') # Is this manually updated?
    args = parser.parse_args()
    the_args = vars(parser.parse_args())

    ### isp output, naive ORF finding:
    isps = OrfFinder(args.table, args.ftype, args.ends, args.isp_min_len, args.strand)
    isps.locate(
        args.fasta_file,
        args.out_isp_nuc,
        args.out_isp_prot,
        args.out_isp_bed,
        args.out_isp_gff3,
    )
    """
    >T7_EIS MLEFLRKLIPWVLVGMLFGLGWHLGSDSMDAKWKQEVHNEYVKRVEAAKSTQRAIGAVSAKYQEDLAALEGSTDRIISDLRSDNKRLRVRVKTTGISDGQCGFEPDGRAELDDRDAKRILAVTQKGDAWIRALQDTIRELQRK
    >lambda_EIS MSRVTAIISALVICIIVCLSWAVNHYRDNAITYKAQRDKNARELKLANAAITDMQMRQRDVAALDAKYTKELADAKAENDALRDDVAAGRRRLHIKAVCQSVREATTASGVDNAASPRLADTAERDYFTLRERLITMQKQLEGTQKYINEQCR
    """
    args.fasta_file.close()
    args.fasta_file = open(args.fasta_file.name, "r")
    args.out_isp_prot.close()
    args.out_isp_prot = open(args.out_isp_prot.name, "r")

    pairs = tuple_fasta(fasta_file=args.out_isp_prot)

    have_tmd = []  # empty candidates list to be passed through the user input criteria

    for each_pair in pairs:
        if len(each_pair[1]) <= args.max_isp:
            try:
                have_tmd += find_tmd(
                    pair=each_pair,
                    minimum=args.isp_min_dist,
                    maximum=args.isp_max_dist,
                    TMDmin=args.min_tmd_size,
                    TMDmax=args.max_tmd_size,
                    isp_mode=args.isp_mode,
                    peri_min=args.peri_min,
                    peri_max=args.peri_max,
                )
            except TypeError:
                continue

    if args.switch == "all":
        pass
    else:
        # for each_pair in have_lipo:
        range_of = args.switch
        range_of = re.search(("[\d]+:[\d]+"), range_of).group(0)
        start = int(range_of.split(":")[0])
        end = int(range_of.split(":")[1])
        have_tmd = parse_a_range(pair=have_tmd, start=start, end=end)

    total_isp = len(have_tmd)

    # ORF = [] # mightttttttttttt use eventually
    length = []  # grabbing length of the sequences
    candidate_dict = {k: v for k, v in have_tmd}
    with args.putative_isp_fa as f:
        for desc, s in candidate_dict.items():  # description / sequence
            f.write(">" + str(desc))
            f.write("\n" + lineWrapper(str(s).replace("*", "")) + "\n")
            length.append(len(s))
            # ORF.append(desc)
    if not length:
        raise Exception("Parameters yielded no candidates.")
    bot_size = min(length)
    top_size = max(length)
    avg = (sum(length)) / total_isp
    n = len(length)
    if n == 0:
        raise Exception("no median for empty data")
    if n % 2 == 1:
        med = length[n // 2]
    else:
        i = n // 2
        med = (length[i - 1] + length[i]) / 2

    #### Extra statistics
    args.out_isp_prot.close()
    all_orfs = open(args.out_isp_prot.name, "r")
    all_isps = open(args.putative_isp_fa.name, "r")
    # record = SeqIO.read(all_orfs, "fasta")
    # print(len(record))
    n = 0
    for line in all_orfs:
        if line.startswith(">"):
            n += 1
    all_orfs_counts = n

    c = 0
    for line in all_isps:
        if line.startswith(">"):
            c += 1
    all_isps_counts = c

    # print(f"{n} -> {c}")
    # count = 0
    # for feature in record.features:
    #    count += 1
    # print(count)

    with args.summary_isp_txt as f:
        f.write("total potential o-spanins: " + str(total_isp) + "\n")
        f.write("average length (AA): " + str(avg) + "\n")
        f.write("median length (AA): " + str(med) + "\n")
        f.write("maximum orf in size (AA): " + str(top_size) + "\n")
        f.write("minimum orf in size (AA): " + str(bot_size) + "\n")
        f.write("ratio of isps found from naive orfs: " + str(c) + "/" + str(n))

    # Output the putative list in gff3 format
    args.putative_isp_fa = open(args.putative_isp_fa.name, "r")
    gff_data = prep_a_gff3(
        fa=args.putative_isp_fa, spanin_type="isp", org=args.fasta_file
    )
    write_gff3(data=gff_data, output=args.putative_isp_gff)

    """https://docs.python.org/3.4/library/subprocess.html"""
    """https://github.tamu.edu/CPT/Galaxy-Tools/blob/f0bf4a4b8e5124d4f3082d21b738dfaa8e1a3cf6/tools/phage/transmembrane.py"""