I used to generate FASTA out of my GenBank source files using a simple conversion script:

#!/usr/bin/env python
from Bio import SeqIO

def wrap( text, width=80 ):
 for i in xrange( 0, len( text ), width ):
  yield text[i:i+width]

if __name__ == "__main__":
  status = progress()
  for record in SeqIO.parse( sys.stdin, "genbank"):
    try:
      gi = record.annotations["gi"]
    except KeyError:
      gi = None
    accession = record.id
    desc = record.description
    seq = record.seq
    locus = record.name
    print ">gi|%s|emb|%s|%s| %s" % (gi, accession, locus, desc)
    for block in wrap( seq ):
      print block

When I changed the sequence files to newer versions some of the resulting FASTA file sequences were just filled with Ns. After closer inspection of the GenBank source files, it turns out that they have replaced the ORIGIN block

ORIGIN
  sequence...

with a CONTIG block, something like

CONTIG      join(BX640437.1:1..347356,BX640438.1:51..347786,...)

It looks like GenBank is now using cross-references to build sequence entries. In my understanding this would require a multi-pass scan of the file. Is there a way to resolve this using BioPython?

Data source: Since the sequences are from the RefSeq collection, I assume they should be self-contained. This problem occurs with the current RefSeq release 47 that you can download via FTP.

I was working with BioPython 1.52 and 1.57 (latest).

Thanks for your suggestions.