Retrieve nonmatching blast queries
The XML output of NCBI's stand alone BLAST programs does not include information on query sequences that have 'no hits' in the target database. Sometimes you want to know which sequences don't have match a database and further analyse/anotate them accordingly. There are a number of different ways to do this, one is to use SeqIO's method .to_dict() to turn the query file into a flat database then parse the results file to get the sequences that did match the database. You can then use python's set() arithmetic to make a list of sequences that are in the query file and not the results which can be used as keys to retrieve the complete SeqRecord for each of the "no hit" queries. Got it? Well, perhaps it's easier to just do it:
Let's presume you set up a BLAST run with the sequences in a file called "queries.fasta" searched against a database, with the results saved to BLAST_RESULTS.xml
from Bio import SeqIO from Bio.Blast import NCBIXML q_dict = SeqIO.to_dict(SeqIO.parse(open('queries.fasta', 'r'), 'fasta')) # make our 'hit list', the 'query' in the BLAST output is the record's ''description'' while # the key for our dictionary is the record id, we need to split the query and get the first # field to get the right key. hits =  for record in NCBIXML.parse(open("BLAST_RESULTS.xml", 'r')): hits.append(record.query.split()) misses = set(q_dict.keys()) - set(hits) orphans =  for record in misses: orphans.append(q_dict[record])
We can do a litte sanity check to make sure everything worked OK:
>>> print "found %i records in query, %i have hits, making %i misses" % (len(q_dict.keys()), len(hits), len(misses)) >>> found 11955 records in query, 2802 have hits, making 9153 misses
Good, now you have all the 'not hits' sequence in a list ('orphans') of SeqRecord objects you can annotate/analyse as you please or use SeqIO.write() to make a new file of just these sequences that can be put through another program.
As implemented above most of the time in each run is spend populating the list of hits from the BLAST parser, would checking each record from the results file against the dictionary one at a time be a less memory intensive way to go in case of very large files?