I got gff3 files (PASA ESTs mapping, exonerate protein2genome mapping) and would like to use it as an Augustus training set. Problem is, Augustus requires Genbank format. Before starting to write my own converter: are there any available already one can recommend?
4 answers
This example script uses Biopython and the in-development python GFF parser. This is essentially the same approach as BioPerl: GFF + Fasta -> generic SeqFeatures -> GenBank. Depending on how nicely the GFF is formatted, this should at least give you a close representation that can be post-processed to exactly what you need:
"""Convert a GFF and associated FASTA file into GenBank format.
Usage:
gff_to_genbank.py <GFF annotation file> <FASTA sequence file>
"""
import sys
import os
from Bio import SeqIO
from Bio.Alphabet import generic_dna
from BCBio import GFF
def main(gff_file, fasta_file):
out_file = "%s.gb" % os.path.splitext(gff_file)[0]
fasta_input = SeqIO.to_dict(SeqIO.parse(fasta_file, "fasta", generic_dna))
gff_iter = GFF.parse(gff_file, fasta_input)
SeqIO.write(gff_iter, out_file, "genbank")
if __name__ == "__main__":
main(*sys.argv[1:])
•
written
by
Brad Chapman
PASA maps ESTs to a whole genome (many contigs). One can limit the size of the file by selecting as fasta input only these contigs to which there is a PASA match. awk/uniq + biopython did the work for me. Going further would be to create some mini-genes containing the matches + some flanks, but that may or may not be desired for Augustus training.
Awesome, glad that worked. Yes, this will require a recent Biopython as the code passes filenames directly to SeqIO; it's a good idea to update anyways to get the latest fixes. Limiting the GFF and fasta files is the right way to go. This is a memory hungry implementation. With some care and GFF files organized by record names, you could build an iterated version that works one record at a time.
AUGUSTUS complains about no 'source' info in Genbank file. Fixed the numbers using biopython. Next had to replace SOURCE with source (Augustus bug), finally one has to have 'CDS' or 'mRNA' for training. Replacing 'cDNA_match' with 'mRNA' does not fix it. Apparently it requires join( (1083..1379, 1503..1595,1865..1930) entry.
Darek -- yes, it will just translate names directly from the original GFF file, so you may have to adjust if they don't match what Augustus wants. It should build join entries if the GFF has nested Parent/Child features. If you post a GFF and fasta example that should help to see what is going on.
Sorry for not answering sooner: viruses got me. Problem is, at this stage all sequence data is still confidential, so I will rerun PASA with some known species next week. While there is a working script in AUGUSTUS (see gvj answer), there is still a place for python solution.
i think this would be a good script to put on the example gff parse page for biopython: http://biopython.org/wiki/GFF_Parsing
Zach, a generalized version of this script is included with the GFF library: https://github.com/chapmanb/bcbb/blob/master/gff/Scripts/gff/gff_to_genbank.py If you'd like to add details to the wiki documentation that would be very welcome. Thanks so much for the feedback.
AUGUSTUS needs a non-protein-coding flanking region around the gene in the genbank file. I don't see this being produced in the python script.
I can't suggest a finished solution, but Bioperl has some components that may help.
There is a GFF3 parser in Bio::Tools::GFF that will allow creation of Bio::SeqFeatureI from GFF3. You'd still need to map the SO terms in the GFF to Genbank.
Bio::SeqFeature::Tools::TypeMapper provides a mapper the other way (Genbank to GFF3), but it can be given a custom mapping. You can probably invert its own mapping and feed it back to it, to make a mapper that goes from SO terms to Genbank feature types. Afterwards, you can use Bio::SeqIO to output Genbank.
•
written
by
biobot 0.0.77.a.1099
Here's an attempt at the solution from the BioPerl mailing list
http://bioperl.org/pipermail/bioperl-l/2005-February/018208.html
Log in to answer this question.
-
Error encountered while initial training with augustus for gene prediction of non model organism
written by lakhujanivijay
-
MAKER annotation gives weird results
written by int11ap1
-
Getting genbank annotation file from JGI
written by bioPiraten •
-
Augustus Training Time
written by Daniel Standage
-
Cufflinks For Rare Transcript Detection
written by Darked89
-
Filtering Spliced Alignment Before Build Gene Models
written by Plantae
-
Ests Cleanup: Seqclean Alternatives?
written by Darked89
-
Genome Annotation Quality Measure
written by Darked89
-
Exonerate Protein2Genome Options And Output
written by Darked89
-
Gene Set(S) For Gene Predictor Training
written by Darked89
