fast, memory-efficient, pythonic (and command-line) access to fasta sequence files
Python Shell
Failed to load latest commit information.
pyfasta info function: don't rely on dict order in doctest Oct 29, 2013
tests code for #16 still needs to handle case where keyfn changes Oct 3, 2013
.gitignore Ignore build, dist directories Apr 30, 2013
README.rst fix str/bytes mismatch Aug 28, 2013
setup.cfg initial git commit. Jul 14, 2010 fix pyfasta extract. thanks @ialbert in #5 . Oct 12, 2011 initial git commit. Jul 14, 2010


pyfasta: pythonic access to fasta sequence files.

NOTE : unless you have specific reason to use pyfasta, I suggest to use pyfaidx

Author:Brent Pedersen (brentp)


Requires Python >= 2.6. Stores a flattened version of the fasta file without spaces or headers and uses either a mmap of numpy binary format or fseek/fread so the sequence data is never read into memory. Saves a pickle (.gdx) of the start, stop (for fseek/mmap) locations of each header in the fasta file for internal use.


>>> from pyfasta import Fasta

>>> f = Fasta('tests/data/three_chrs.fasta')
>>> sorted(f.keys())
['chr1', 'chr2', 'chr3']

>>> f['chr1']


# get full the sequence:
>>> a = str(f['chr1'])
>>> b = f['chr1'][:]
>>> a == b

>>> f['chr1'][:10]

# get the 1st basepair in every codon (it's python yo)
>>> f['chr1'][::3]

# can query by a 'feature' dictionary (note this is one based coordinates)
>>> f.sequence({'chr': 'chr1', 'start': 2, 'stop': 9})

# same as:
>>> f['chr1'][1:9]

# use python, zero based coords
>>> f.sequence({'chr': 'chr1', 'start': 2, 'stop': 9}, one_based=False)

# with reverse complement (automatic for - strand)
>>> f.sequence({'chr': 'chr1', 'start': 2, 'stop': 9, 'strand': '-'})

Key Function

Sometimes your fasta will have a long header like: "AT1G51370.2 | Symbols: | F-box family protein | chr1:19045615-19046748 FORWARD" when you only want to key off: "AT1G51370.2". In this case, specify the key_fn argument to the constructor:

>>> fkey = Fasta('tests/data/key.fasta', key_fn=lambda key: key.split()[0])
>>> sorted(fkey.keys())
['a', 'b', 'c']


The default is to use a memmaped numpy array as the backend. In which case it's possible to get back an array directly...

>>> f['chr1'].as_string = False
>>> f['chr1'][:10] # doctest: +NORMALIZE_WHITESPACE
memmap(['A', 'C', 'T', 'G', 'A', 'C', 'T', 'G', 'A', 'C'], dtype='|S1')

>>> import numpy as np
>>> a = np.array(f['chr2'])
>>> a.shape[0] == len(f['chr2'])

>>> a[10:14] # doctest: +NORMALIZE_WHITESPACE
array(['A', 'A', 'A', 'A'], dtype='|S1')

mask a sub-sequence

>>> a[11:13] = np.array('N', dtype='S1')
>>> a[10:14].tostring()

Backends (Record class)

It's also possible to specify another record class as the underlying work-horse for slicing and reading. Currently, there's just the default:

  • NpyFastaRecord which uses numpy memmap
  • FastaRecord, which uses using fseek/fread
  • MemoryRecord which reads everything into memory and must reparse the original fasta every time.
  • TCRecord which is identical to NpyFastaRecord except that it saves the index in a TokyoCabinet hash database, for cases when there are enough records that loading the entire index from a pickle into memory is unwise. (NOTE: that the sequence is not loaded into memory in either case).

It's possible to specify the class used with the record_class kwarg to the Fasta constructor:

>>> from pyfasta import FastaRecord # default is NpyFastaRecord
>>> f = Fasta('tests/data/three_chrs.fasta', record_class=FastaRecord)
>>> f['chr1']
FastaRecord('tests/data/three_chrs.fasta.flat', 0..80)

other than the repr, it should behave exactly like the Npy record class backend

it's possible to create your own using a sub-class of FastaRecord. see the source in pyfasta/ for details.


In order to efficiently access the sequence content, pyfasta saves a separate, flattened file with all newlines and headers removed from the sequence. In the case of large fasta files, one may not wish to save 2 copies of a 5GG+ file. In that case, it's possible to flatten the file "inplace", keeping all the headers, and retaining the validity of the fasta file -- with the only change being that the new-lines are removed from each sequence. This can be specified via flatten_inplace = True

>>> import os
>>> os.unlink('tests/data/three_chrs.fasta.gdx') # cleanup non-inplace idx
>>> f = Fasta('tests/data/three_chrs.fasta', flatten_inplace=True)
>>> f['chr1']  # note the difference in the output from above.

# sequence from is same as when requested from non-flat file above.
>>> f['chr1'][1:9]

# the flattened file is kept as a place holder without the sequence data.
>>> open('tests/data/three_chrs.fasta.flat').read()

Command Line Interface

there's also a command line interface to manipulate / view fasta files. the pyfasta executable is installed via setuptools, running it will show help text.

split a fasta file into 6 new files of relatively even size:

$ pyfasta split -n 6 original.fasta

split the fasta file into one new file per header with "%(seqid)s" being filled into each filename.:

$ pyfasta split --header "%(seqid)s.fasta" original.fasta

create 1 new fasta file with the sequence split into 10K-mers:

$ pyfasta split -n 1 -k 10000 original.fasta

2 new fasta files with the sequence split into 10K-mers with 2K overlap:

$ pyfasta split -n 2 -k 10000 -o 2000 original.fasta

show some info about the file (and show gc content):

$ pyfasta info --gc test/data/three_chrs.fasta

extract sequence from the file. use the header flag to make a new fasta file. the args are a list of sequences to extract.

$ pyfasta extract --header --fasta test/data/three_chrs.fasta seqa seqb seqc

extract sequence from a file using a file containing the headers not wanted in the new file:

$ pyfasta extract --header --fasta input.fasta --exclude --file seqids_to_exclude.txt

extract sequence from a fasta file with complex keys where we only want to lookup based on the part before the space.

$ pyfasta extract --header --fasta input.with.keys.fasta --space --file seqids.txt

flatten a file inplace, for faster later use by pyfasta, and without creating another copy. (Flattening)

$ pyfasta flatten input.fasta


(though for real use these will remain for faster access)

>>> os.unlink('tests/data/three_chrs.fasta.gdx')
>>> os.unlink('tests/data/three_chrs.fasta.flat')


there is currently > 99% test coverage for the 2 modules and all included record classes. to run the tests:

$ python nosetests