Skip to content
non-redundant, compressed, journalled, file-based storage for biological sequences
Python Makefile Shell Perl
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.



Python package for writing and reading a local collection of biological sequences. The repository is non-redundant, compressed, and journalled, making it efficient to store and transfer multiple snapshots.

Clients refer to sequences and metadata using familiar identifiers, such as NM_000551.3 or GRCh38:1, or any of several hash-based identifiers. The interface supports fast slicing of arbitrary regions of large sequences.

A "fully-qualified" identifier includes a namespace to disambiguate accessions (e.g., "1" in GRCh37 and GRCh38). If the namespace is provided, seqrepo uses it as-is. If the namespace is not provided and the unqualified identifier refers to a unique sequence, it is returned; otherwise, ambiguous identifiers will raise an error.

seqrepo-rest-service provides a REST interface and docker image.

Released under the Apache License, 2.0.

ci_rel | cov | pypi_rel


  • Timestamped, read-only snapshots.
  • Space-efficient storage of sequences within a single snapshot and across snapshots.
  • Bandwidth-efficient transfer incremental updates.
  • Fast fetching of sequence slices on chromosome-scale sequences.
  • Precomputed digests that may be used as sequence aliases.
  • Mappings of external aliases (i.e., accessions or identifiers like NM_013305.4) to sequences.

Deployments Scenarios

  • Local read-only archive, mirrored from public site, accessed via Python API (see Mirroring documentation)
  • Local read-write archive, maintained with command line utility and/or API (see Command Line Interface documentation).
  • Docker-based data-only container that may be linked to application container.
  • Planned: Docker image that provides REST interface for local or remote access

Technical Quick Peek

Within a single snapshot, sequences are stored non-redundantly and compressed in an add-only journalled filesystem structure. A truncated SHA-512 hash is used to assess uniquness and as an internal id. (The digest is truncated for space efficiency.)

Sequences are compressed using the Block GZipped Format (BGZF)), which enables pysam to provide fast random access to compressed sequences. (Variable compression typically makes random access impossible.)

Sequence files are immutable, thereby enabling the use of hardlinks across snapshots and eliminating redundant transfers (e.g., with rsync).

Each sequence id is associated with a namespaced alias in a sqlite database. Such as <seguid,rvvuhY0FxFLNwf10FXFIrSQ7AvQ>, <NCBI,NP_004009.1>, <gi,5032303>, <ensembl-75ENSP00000354464>, <ensembl-85,ENSP00000354464.4>. The sqlite database is mutable across releases.

For calibration, recent releases that include 3 human genome assemblies (including patches), and full RefSeq sets (NM, NR, NP, NT, XM, and XP) consumes approximately 8GB. The minimum marginal size for additional snapshots is approximately 2GB (for the sqlite database, which is not hardlinked).

For more information, see docs/design.rst.


Reading a sequence repository requires several Python packages, all of which are available from pypi. Installation should be as simple as pip install biocommons.seqrepo.

Writing sequence files also requires bgzip, which provided in the htslib repo. Ubuntu users should install the tabix package with sudo apt install tabix.

Development and deployments are on Ubuntu. Other systems may work but are not tested. Patches to get other systems working would be welcomed.

Mac Developers If you get "xcrun: error: invalid active developer path", you need to install XCode. See this StackOverflow answer.

Quick Start

On Ubuntu 16.04:

$ sudo apt install -y python3-dev gcc zlib1g-dev tabix
$ pip install seqrepo
$ sudo mkdir /usr/local/share/seqrepo
$ sudo chown $USER /usr/local/share/seqrepo
$ seqrepo pull -i 2018-11-26
$ seqrepo show-status -i 2018-11-26
seqrepo 0.2.3.post3.dev8+nb8298bd62283
root directory: /usr/local/share/seqrepo/2018-11-26, 7.9 GB
backends: fastadir (schema 1), seqaliasdb (schema 1)
sequences: 773587 sequences, 93051609959 residues, 192 files
aliases: 5579572 aliases, 5480085 current, 26 namespaces, 773587 sequences

# Simple Pythonic interface to sequences
>> from biocommons.seqrepo import SeqRepo
>> sr = SeqRepo("/usr/local/share/seqrepo/latest")
>> sr["NC_000001.11"][780000:780020]

# Or, use the seqrepo shell for even easier access
$ seqrepo start-shell -i 2018-11-26
In [1]: sr["NC_000001.11"][780000:780020]

# N.B. The following output is edited for simplicity
$ seqrepo export -i 2018-11-26 | head -n100
>SHA1:9a2acba3dd7603f... SEGUID:mirLo912A/MppLuS1cUyFMduLUQ Ensembl-85:GENSCAN00000003538 ...
>SHA1:ca996b263102b1... SEGUID:yplrJjECsVqQufeYy0HkDD16z58 NCBI:XR_001733142.1 gi:1034683989

See Installation and Mirroring for more information.


Here's how to get started developing:

python3.6 -m venv
source venv/bin/activate
pip install -U setuptools pip
make develop
You can’t perform that action at this time.