skip to navigation
skip to content

Not Logged In

vdw 0.4.1

Models and loader for processing, storing, and querying genetic data

# Varify Data Warehouse Development Guide

[![Build Status](]( [![Coverage Status](](

## Need some help?
Join our chat room and speak with our dev team:

## Dependencies

Listed are the download links to each dependency, however most OSes have a
package manager or binaries that can be easily installed. Most of the below
links describe alternate download and install methods.

On Mac OS X, [Homebrew]( is the recommended
way to install most of these libraries.

- [Python 2.6+](
- [Redis 2.6+](
- [PostgreSQL 9.2+](
- [Memcached](

## Setup & Install

Distribute, Pip and virtualenv are required. To check if you have them:

which pip easy_install virtualenv

If nothing prints out, install the libraries corresponding to the commands

_Watch out for sudo! The root user `$PATH` most likely does not include
`/usr/local/bin`. If you did not install Python through your distro's package
manager, use the absolute path to the new Python binary to prevent installing
the above libraries with the wrong version (like Python 2.4 on CentOS 5),
e.g. `/usr/local/bin/python2.7`._

curl | python
curl | python
pip install virtualenv

Create your virtualenv:

virtualenv vdw-env
cd vdw-env
. bin/activate

Clone the repo:

git clone
cd varify-data-warehouse

Install the requirements:

pip install -r requirements.txt

Under Mac OS X 10.8 or later, with XCode 5.1 or later, the following may be necessary in order for pip to install requirements:

export CFLAGS=-Qunused-arguments

[Start the postgres server]( This *may* look something like:
initdb /usr/local/var/postgres -E utf8

pg_ctl -D /usr/local/var/postgres -l /usr/local/var/postgres/server.log start

Create the varify database, you might first want to make sure you are a user
createuser --user postgres -s -r yourusername
createdb varify


Start memcached
memcached -d

Start redis
redis-server /usr/local/etc/redis.conf

If you are on a Mac, you will need to start postfix to allow SMTP:
sudo postfix start

Initialize the Django and Varify schemas
./bin/ syncdb
./bin/ migrate

Then either start the built-in Django server:

./bin/ runserver

or run a `uwsgi` process:

uwsgi --ini server/uwsgi/local.ini --protocol http --socket --check-static _site

## Local Settings

`` is intentionally not versioned (via `.gitignore`). It should
contain any environment-specific settings and/or sensitive settings such as
passwords, the `SECRET_KEY` and other information that should not be in version
control. Defining `` is not mandatory but will warn if it does
not exist.

## Pipeline

The following describes the steps to execute the loading pipeline, the performance of the pipeline, and the process behind it.

#### NOTE: All VCF files being loaded into Varify must be annotated with the [CBMi fork of SnpEff]( The key difference is that the CBMi fork attempts to generate valid HGVS for insertions and deletions, including those which require "walking and rolling" to identify the correct indel frame while the standard SnpEff version only contains a partial implementation of HGVS notation [as noted here](

### Retrieving Test Data

We have provided a [set of test data]( to use to test the load pipeline or use as sample data when first standing up your Varify instance. To use the test data, run the commands below.

wget -O varify-demo-data-0.1.tar.gz
tar -zxf varify-demo-data-0.1.tar.gz
gunzip varify-demo-data-0.1/CEU.trio.2010_03.genotypes.annotated.vcf.gz

At this point, the VCF and MANIFEST in the `varify-demo-data-0.1` directory are ready for loading in the pipeline. You can use the `varify-demo-data-0.1` directory as the argument to the `samples queue` command in the _Queue Samples_ step below if you want to just load this test data.

#### Tmux (optional)

Since the pipeline can take a while to load large collections(see Performance section below), you may want to consider following the Tmux steps to attach/detach to/from the load process.

[Tmux]( is like [screen]( just newer. It is useful for detaching/reattaching sessions with long running processes.

**New Session**


**Existing Session**

tmux attach -t 0 # first session

### Activate Environment

source bin/activate

#### Define RQ_QUEUES

For this example, we will assume you have `redis-server` running on `localhost:6379` against the database with index 0. If you have redis running elsewhere simply update the settings below with the address info and DB you wish to use. Open your `` file and add the following setting:

'default': {
'HOST': 'localhost',
'PORT': 6379,
'DB': 0,
'samples': {
'HOST': 'localhost',
'PORT': 6379,
'DB': 0,
'variants': {
'HOST': 'localhost',
'PORT': 6379,
'DB': 0,

#### Queue Samples

Optionally specify a directory, otherwise it will recursively scan all directories defined in the `VARIFY_SAMPLE_DIRS` setting in the Varify project.

./bin/ samples queue [directory]

#### Kick Off Workers

You can technically start as many of each type for loading data in parallel, but this may cause undesired database contention which could actually slow down the loading process. A single worker for `variants` is generally preferred and two or three are suitable for the `default` type.

./bin/ rqworker variants &
./bin/ rqworker default &

Note, these workers will run forever, if there is only a single sample being loaded, the `--burst` argument can be used to terminate the worker when there are no more items left in the queue.

#### Monitor Workers

You can monitor the workers and the queues using the `rq-dashboard` or `rqinfo`. Information on setting up and using those services can be found [here](

#### Post-Load

After the batch of samples have been loaded, two more commands need to be executed to update the annotations and cohort frequencies. These are performed _post-load_ for performance reasons.

./bin/ variants load --evs --1000g --sift --polyphen2 > variants.load.txt 2>&1 &
./bin/ samples allele-freqs > samples.allele-freqs.txt 2>&1 &

### Performance

- File size: 610 MB
- Variant count: 1,794,055

#### Baseline

Iteration over flat file (no parsing) with batch counting (every 1000)

- Time: 80 seconds
- Memory: 0

#### Baseline VCF

Iteration over VCF parsed file using PyVCF

- Time: 41 minutes (extrapolated)
- Memory: 246 KB

### Parallelized Queue/Worker Process

#### Summary of Workflow

1. Fill Queue
2. Spawn Worker(s)
3. Consume Job(s)
- Validate Input
- (work)
- Validate Output
- Commit

#### Parallelism Constraints

The COPY command is a single statement which means the data being loaded is
all or nothing. If multiple samples are being loaded in parallel, it is likely
they will have overlapping variants.

To prevent integrity errors, workers will need to consult one or more
centralized caches to check if the current variant has been _addressed_
already. If this is the case, the variant will be skipped by the worker.

This incurs a second issue in that downstream jobs depend on the existence of
some data that does not yet exist because another worker has not yet committed
its data. In this case, non-matches will be queued up in the `deferred` queue
that can be run at a later time, after the `default` queue is empty or in
parallel with the `default` queue.

File Type Py Version Uploaded on Size
vdw-0.4.1.tar.gz (md5) Source 2014-10-17 112KB
  • Downloads (All Versions):
  • 7 downloads in the last day
  • 110 downloads in the last week
  • 381 downloads in the last month