Mercurial > repos > cpt > cpt_gbk_adjacent
view adjacent_features.py @ 1:e29c36ee61e0 draft
planemo upload commit 94b0cd1fff0826c6db3e7dc0c91c0c5a8be8bb0c
author | cpt |
---|---|
date | Mon, 05 Jun 2023 02:42:35 +0000 |
parents | |
children |
line wrap: on
line source
#!/usr/bin/env python from Bio import SeqIO from Bio.Seq import Seq from Bio.Data import CodonTable from Bio.SeqRecord import SeqRecord from Bio.SeqFeature import SeqFeature, FeatureLocation from Bio.Alphabet import generic_dna, generic_protein import argparse import logging logging.basicConfig(level=logging.INFO) log = logging.getLogger() def extract_features( genbankFiles=None, fastaFiles=None, upOut=None, downOut=None, genesOnly=False, cdsOnly=True, forceSeqID=False, forward=1, behind=1, outProt=True, tTable=11, fTable=11, ): genList = [] fastaList = [] for fileX in genbankFiles: opener = SeqIO.parse(fileX, "genbank") for ( openRec ) in ( opener ): # To turn the generator into objects (or else we end up with a list of generators) genList.append(openRec) for fileX in fastaFiles: opener = SeqIO.parse(fileX, "fasta") for openRec in opener: # Technically flattens multifastas too fastaList.append(openRec) for seqMatch in fastaList: longOut = seqMatch.description protID = seqMatch.id if fTable != 0: fSeq = seqMatch.seq.translate(table=fTable, cds=False) else: fSeq = seqMatch.seq for gbk in genList: sourceOut = gbk.id num = -1 for feat in gbk.features: num += 1 if (genesOnly and feat.type != "gene") or ( cdsOnly and feat.type != "CDS" ): continue if "codon_start" in feat.qualifiers: offset = 1 - int(feat.qualifiers["codon_start"][0]) else: offset = 0 temp = gbk.seq[feat.location.start : feat.location.end] if feat.location.strand == -1: temp = gbk.seq[feat.location.start : feat.location.end - offset] temp = temp.reverse_complement() else: temp = gbk.seq[feat.location.start + offset : feat.location.end] if tTable != 0: try: gSeq = temp.translate(table=tTable, cds=True) except CodonTable.TranslationError as cte: # log.info("Translation issue at %s", cte) gSeq = temp.translate(table=tTable, cds=False) else: gSeq = temp if not ("protein_id" in feat.qualifiers): feat.qualifiers["protein_id"] = [ "++++++++" ] # Junk value for genesOnly flag if (gSeq == fSeq) and ( protID == feat.qualifiers["protein_id"][0] or forceSeqID == False ): goBack = num - 1 goAhead = num + 1 numBack = behind numAhead = forward backList = [] aheadList = [] while numBack != 0 and goBack >= 0: if (genesOnly and gbk.features[goBack].type != "gene") or ( cdsOnly and gbk.features[goBack].type != "CDS" ): goBack -= 1 continue backList.append(gbk.features[goBack]) numBack -= 1 goBack -= 1 while numAhead != 0 and goAhead < len(gbk.features): if (genesOnly and gbk.features[goAhead].type != "gene") or ( cdsOnly and gbk.features[goAhead].type != "CDS" ): goAhead += 1 continue aheadList.append(gbk.features[goAhead]) numAhead -= 1 goAhead += 1 backList.reverse() if feat.location.strand == -1: tmpList = aheadList aheadList = backList backList = tmpList for item in backList: addition = "" header = "" if "product" in item.qualifiers: addition = " -" + str(item.qualifiers["product"][0]) + "-" if "protein_id" in item.qualifiers: header = ( ">" + (item.qualifiers["protein_id"][0]) + addition + " (5' of " + longOut + " found within " + sourceOut + ")\n" ) else: header = ( ">" + (item.qualifiers["locus_tag"][0]) + addition + " (5' of " + longOut + " found within " + sourceOut + ")\n" ) if outProt == True: if "translation" in item.qualifiers: upOut.write(header) upOut.write( str(item.qualifiers["translation"][0]) + "\n\n" ) else: modS = 0 modE = 0 if "codon_start" in item.qualifiers: if item.location.strand > 0: modS = ( int(item.qualifiers["codon_start"][0]) - 1 ) else: modE = ( int(item.qualifiers["codon_start"][0]) - 1 ) seqHold = gbk.seq[ item.location.start + modS : item.location.end - modE ] if item.location.strand == -1: seqHold = seqHold.reverse_complement() if cdsOnly: try: finalSeq = "" if tTable != 0: finalSeq = ( str( seqHold.translate( table=tTable, cds=True ) ) + "\n\n" ) else: finalSeq = str(seqHold) + "\n\n" # upOut.write(header) # upOut.write(finalSeq) except Exception as bdct: log.warn( "ERROR %s %s", item.qualifiers["locus_tag"][0], bdct, ) finalSeq = "" if tTable != 0: finalSeq = ( str( seqHold.translate( table=tTable, cds=False ) ) + "\n\n" ) else: finalSeq = str(seqHold) + "\n\n" header = ( ">" + (item.qualifiers["locus_tag"][0]) + addition + " [INCOMPLETE] (5' of " + longOut + " found within " + sourceOut + ")\n" ) upOut.write(header) upOut.write(finalSeq) else: if tTable != 0: upOut.write(header) upOut.write( str( seqHold.translate( table=tTable, cds=False ) ) + "\n\n" ) else: upOut.write(header) upOut.write(str(seqHold) + "\n\n") else: upOut.write(header) upOut.write( str(gbk.seq[item.location.start : item.location.end]) + "\n\n" ) for item in aheadList: addition = "" header = "" if "product" in item.qualifiers: addition = " -" + str(item.qualifiers["product"][0]) + "-" if "protein_id" in item.qualifiers: header = ( ">" + (item.qualifiers["protein_id"][0]) + addition + " (3' of " + longOut + " found within " + sourceOut + ")\n" ) else: header = ( ">" + (item.qualifiers["locus_tag"][0]) + addition + " (3' of " + longOut + " found within " + sourceOut + ")\n" ) if outProt == True: if "translation" in item.qualifiers: downOut.write(header) downOut.write( str(item.qualifiers["translation"][0]) + "\n\n" ) else: modS = 0 modE = 0 if "codon_start" in item.qualifiers: if item.location.strand > 0: modS = ( int(item.qualifiers["codon_start"][0]) - 1 ) else: modE = ( int(item.qualifiers["codon_start"][0]) - 1 ) seqHold = gbk.seq[ item.location.start + modS : item.location.end - modE ] if item.location.strand == -1: seqHold = seqHold.reverse_complement() if cdsOnly: try: finalSeq = "" if tTable != 0: finalSeq = ( str( seqHold.translate( table=tTable, cds=True ) ) + "\n\n" ) else: finalSeq = str(seqHold) + "\n\n" # downOut.write(header) # downOut.write(finalSeq) except Exception as bdct: log.warn( "ERROR %s %s", item.qualifiers["locus_tag"][0], bdct, ) finalSeq = "" if tTable != 0: finalSeq = ( str( seqHold.translate( table=tTable, cds=False ) ) + "\n\n" ) else: finalSeq = str(seqHold) + "\n\n" header = ( ">" + (item.qualifiers["locus_tag"][0]) + addition + " [INCOMPLETE] (3' of " + longOut + " found within " + sourceOut + ")\n" ) downOut.write(header) downOut.write(finalSeq) else: if tTable != 0: downOut.write(header) downOut.write( str( seqHold.translate( table=tTable, cds=False ) ) + "\n\n" ) else: downOut.write(header) downOut.write(str(seqHold) + "\n\n") else: downOut.write(header) downOut.write( str(gbk.seq[item.location.start : item.location.end]) + "\n\n" ) # print(longOut) return if __name__ == "__main__": parser = argparse.ArgumentParser( description="Export a subset of features from a Genbank file", epilog="" ) parser.add_argument( "-genbankFiles", nargs="+", type=argparse.FileType("r"), help="Genbank file" ) parser.add_argument( "-fastaFiles", nargs="+", type=argparse.FileType("r"), help="Fasta file to match against", ) parser.add_argument( "-tTable", type=int, default=11, help="Translation table to use", choices=range(0, 23), ) parser.add_argument( "-fTable", type=int, default=11, help="Translation table to use", choices=range(0, 23), ) parser.add_argument( "-upOut", type=argparse.FileType("w"), help="Upstream Fasta output", default="test-data/upOut.fa", ) parser.add_argument( "-downOut", type=argparse.FileType("w"), help="Downstream Fasta output", default="test-data/downOut.fa", ) parser.add_argument( "--genesOnly", action="store_true", help="Search and return only Gene type features", ) parser.add_argument( "--cdsOnly", action="store_true", help="Search and return only CDS type features", ) parser.add_argument( "--forceSeqID", action="store_true", help="Search and return only CDS type features", ) parser.add_argument( "--outProt", action="store_true", help="Output the translated sequence" ) parser.add_argument( "--forward", type=int, default=1, help="Number of features upstream from the hit to return", ) parser.add_argument( "--behind", type=int, default=1, help="Number of features downstream from the hit to return", ) args = parser.parse_args() extract_features(**vars(args))