Skip to content

non-redundant, compressed, journalled, file-based storage for biological sequences

License

Notifications You must be signed in to change notification settings

biocommons/biocommons.seqrepo

Repository files navigation

biocommons.seqrepo

SeqRepo is a Python package for storing 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.

Introduction

Specific, named biological sequences provide the reference and coordinate system for communicating variation and consequential phenotypic changes. Several databases of sequences exist, with significant overlap, all using distinct names. Furthermore, these systems are often difficult to install locally.

SeqRepo provides an efficient, non-redundant and indexed storage system for biological sequences. 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 from different origins or sequence sets (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, the use of ambiguous identifiers raise an error.

SeqRepo favors namespaces from identifiers.org whenever available. Examples include refseq and ensembl.

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

Released under the Apache License, 2.0.

ci_rel | cov | pypi_rel | ChangeLog

Citation

Hart RK, Prlić A (2020). SeqRepo: A system for managing local collections of biological sequences. PLoS ONE 15(12): e0239883. https://doi.org/10.1371/journal.pone.0239883

Features

  • 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

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.

Requirements

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.

Quick Start

OS X

$ brew install python libpq

Ubuntu

$ sudo apt install -y python3-dev gcc zlib1g-dev tabix

All platforms

$ python -m venv venv
$ source venv/bin/activate
$ pip install seqrepo
$ sudo mkdir -p /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]
'TGGTGGCACGCGCTTGTAGT'

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

# 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 ...
MDSPLREDDSQTCARLWEAEVKRHSLEGLTVFGTAVQIHNVQRRAIRAKGTQEAQAELLCRGPRLLDRFLEDACILKEGRGTDTGQHCRGDARISSHLEA
SGTHIQLLALFLVSSSDTPPSLLRFCHALEHDIRYNSSFDSYYPLSPHSRHNDDLQTPSSHLGYIITVPDPTLPLTFASLYLGMAPCTSMGSSSMGIFQS
QRIHAFMKGKNKWDEYEGRKESWKIRSNSQTGEPTF
>SHA1:ca996b263102b1... SEGUID:yplrJjECsVqQufeYy0HkDD16z58 NCBI:XR_001733142.1 gi:1034683989
TTTACGTCTTTCTGGGAATTTATACTGGAAGTATACTTACCTCTGTGCAAAATTGCAAATATATAAGGTAATTCATTCCAGCATTGCTTATATTAGGTTG
AACTATGTAACATTGACATTGATGTGAATCAAAAATGGTTGAAGGCTGGCAGTTTCATATGATTCAGCCTATAATAGCAAAAGATTGAAAAAATCCATTA
ATACAGTGTGGTTCAAAAAAATTTGTTGTATCAAGGTAAAATAATAGCCTGAATATAATTAAGATAGTCTGTGTATACATCGATGAAAACATTGCCAATA

See Installation and Mirroring for more information.

Environment Variables

SEQREPO_LRU_CACHE_MAXSIZE sets the lru_cache maxsize for the sqlite query response caching. It defaults to 1 million but can also be set to "none" to be unlimited.

SEQREPO_FD_CACHE_MAXSIZE sets the lru_cache size for file handler caching during FASTA sequence retrievals. It defaults to 0 to disable any caching, but can be set to a specific value or "none" to be unlimited. Using a moderate value (>10) will greatly increase performance of sequence retrieval.

Developing

Developing on OS X

brew install python libpq bash

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

Developing on Ubuntu

sudo apt install -y python3-dev gcc zlib1g-dev tabix

Here's how to get started developing:

make devready
source venv/bin/activate
seqrepo --version

Code reformatting:

make reformat

Install pre-commit hook:

# included in `make devready`, not necessary for new installations
pre-commit install

Building a docker image

Docker images are available at https://hub.docker.com/r/biocommons/seqrepo. Tags correspond to the version of data, not the version of seqrepo, because the intent is to make it easy to depend on a local version of seqrepo files. Each docker image is an installation of seqrepo that downloads the corresponding version of seqrepo data. When used in conjunction with docker volumes for persistence, this provides an easy way to incorporate seqrepo data into a docker stack.

Building

cd misc/docker
make 2021-01-29.log  # builds and pushes to hub.docker.com (i.e., you need creds)