Querying the Metadata Database


The humans, specimens and traits tables are deprecated and will be removed in a future release. The recommended way to store and search on user-defined metadata is using the “properties” field of Arvados resources.

This tutorial introduces the Arvados Metadata Database. The Metadata Database stores information about files in Keep. This example will use the Python SDK to find public WGS (Whole Genome Sequencing) data for people who have reported a certain medical condition.


This tutorial assumes that you are logged into an Arvados VM instance (instructions for Webshell or Unix or Windows) or you have installed the Arvados FUSE Driver and Python SDK on your workstation and have a working environment.

In the tutorial examples, three angle brackets (>>>) will be used to denote code to enter at the interactive Python prompt.

Start by running Python.

~$ python
Python 2.7.3 (default, Jan  2 2013, 13:56:14)
[GCC 4.7.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.

If everything is set up correctly, you will be able to import the arvados SDK.

>>> import arvados

This tutorial will also use the regular expression (re) python module:

>>> import re

Finding traits

>>> all_traits = arvados.api().traits().list(limit=1000).execute()
  • arvados.api() gets an object that provides access to the Arvados API server
  • .traits() gets an object that provides access to the “traits” resource on the Arvados API server
  • .list(limit=1000) constructs a query to list all elements of the “traits” resource, with a limit of 1000 entries returned
  • .execute() executes the query and returns the result, which we assign to “all_traits”
>>> cancer_traits = filter(lambda t: re.search('cancer', t['name']), all_traits['items'])
  • lambda t: re.search('cancer', t['name']) is an inline function that takes a parameter t and uses a simple regular expression to test if t['name'] contains the substring ‘cancer’
  • all_traits['items'] is the input sequence of traits
  • filter tests each element t and constructs a new sequence consisting only of the elements that pass the filter
  • cancer_traits gets the result of filter
>>> for t in cancer_traits: print(t['uuid'], t['name'])
qr1hi-q1cn2-8q57g2diohwnzm0 Cervical cancer
qr1hi-q1cn2-vqp4243janpjbyj Breast cancer
qr1hi-q1cn2-v6usijujcpwqrn1 Non-melanoma skin cancer

In this tutorial wil will use “Non-melanoma skin cancer” trait with uuid qr1hi-q1cn2-v6usijujcpwqrn1.

>>> non_melanoma_cancer = 'qr1hi-q1cn2-v6usijujcpwqrn1'

Finding humans with the selected trait

We query the “links” resource to find humans that report the selected trait. Links are directional connections between Arvados data items, for example, from a human to their reported traits.

>>> trait_filter = [
    ['link_class', '=', 'human_trait'],
    ['tail_uuid', 'is_a', 'arvados#human'],
    ['head_uuid', '=', non_melanoma_cancer],
  • ['link_class', '=', 'human_trait'] filters on links that connect phenotype traits to individuals in the database.
  • ['tail_uuid', 'is_a', 'arvados#human'] filters that the “tail” must be a “human” database object.
  • ['head_uuid', '=', non_melanoma_cancer] filters that the “head” of the link must connect to the “trait” database object non_melanoma_cancer .

The query will return links that match all three conditions.

>>> trait_links = arvados.api().links().list(limit=1000, filters=trait_filter).execute()
  • arvados.api() gets an object that provides access to the Arvados API server
  • .links() gets an object that provides access to the “links” resource on the Arvados API server
  • .list(limit=1000, filters=trait_filter) constructs a query to elements of the “links” resource that match the criteria discussed above, with a limit of 1000 entries returned
  • .execute() executes the query and returns the result, which we assign to “trait_links”
>>> human_uuids = map(lambda l: l['tail_uuid'], trait_links['items'])
>>> human_uuids
[u'1h9kt-7a9it-c0uqa4kcdh29wdf', u'1h9kt-7a9it-x4tru6mn40hc6ah',
u'1h9kt-7a9it-yqb8m5s9cpy88i8', u'1h9kt-7a9it-46sm75w200ngwny',
u'1h9kt-7a9it-gx85a4tdkpzsg3w', u'1h9kt-7a9it-8cvlaa8909lgeo9',
u'1h9kt-7a9it-as37qum2pq8vizb', u'1h9kt-7a9it-14fph66z2baqxb9',
u'1h9kt-7a9it-e9zc7i4crmw3v69', u'1h9kt-7a9it-np7f35hlijlxdmt',
u'1h9kt-7a9it-j9hqyjwbvo9cojn', u'1h9kt-7a9it-lqxdtm1gynmsv13',
u'1h9kt-7a9it-zkhhxjfg2o22ywq', u'1h9kt-7a9it-nsjoxqd33lzldw9',
u'1h9kt-7a9it-ytect4smzcgd4kg', u'1h9kt-7a9it-y6tl353b3jc4tos',
u'1h9kt-7a9it-98f8qave4f8vbs5', u'1h9kt-7a9it-gd72sh15q0p4wq3',
u'1h9kt-7a9it-zlx25dscak94q9h', u'1h9kt-7a9it-8gronw4rbgmim01',
u'1h9kt-7a9it-wclfkjcb23tr5es', u'1h9kt-7a9it-rvp2qe7szfz4dy6',
u'1h9kt-7a9it-50iffhmpzsktwjm', u'1h9kt-7a9it-ul412id5y31a5o8',
u'1h9kt-7a9it-732kwkfzylmt4ik', u'1h9kt-7a9it-v9zqxegpblsbtai',
u'1h9kt-7a9it-kmaraqduit1v5wd', u'1h9kt-7a9it-t1nwtlo1hru5vvq',
u'1h9kt-7a9it-q3w6j9od4ibpoyl', u'1h9kt-7a9it-qz8vzkuuz97ezwv',
u'1h9kt-7a9it-t1v8sjz6dm9jmjf', u'1h9kt-7a9it-qe8wrbyvuqs5jew']
  • lambda l: l['tail_uuid'] is an inline function that returns the ‘tail_uuid’ attribute of ‘l’
  • trait_links['items'] is the input set from the query
  • map converts each item in a sequence into a different item using the embedded function, in this case to produce a sequence of uuids which refer to humans that have the specified trait.

Find Personal Genome Project identifiers from Arvados UUIDs

>>> human_filters = [
    ["link_class", "=", "identifier"],
    ["head_uuid", "in", human_uuids]
>>> pgpid_links = arvados.api('v1').links().list(limit=1000, filters=human_filters).execute()
>>> map(lambda l: l['name'], pgpid_links['items'])
[u'hu01024B', u'hu11603C', u'hu15402B', u'hu174334', u'hu1BD549', u'hu237A50',
 u'hu34A921', u'hu397733', u'hu414115', u'hu43860C', u'hu474789', u'hu553620',
 u'hu56B3B6', u'hu5917F3', u'hu599905', u'hu5E55F5', u'hu602487', u'hu633787',
 u'hu68F245', u'hu6C3F34', u'hu7260DD', u'hu7A2F1D', u'hu94040B', u'hu9E356F',
 u'huAB8707', u'huB1FD55', u'huB4883B', u'huD09050', u'huD09534', u'huD3A569',
 u'huDF04CC', u'huE2E371']

These PGP IDs let us find public profiles, for example:

Find genomic data from specific humans

Now we want to find collections in Keep that were provided by these humans. We search the “links” resource for “provenance” links that point to entries in the list of humans with the non-melanoma skin cancer trait:

>>> provenance_links = arvados.api().links().list(limit=1000, filters=[
    ["link_class", "=", "provenance"],
    ["name", "=", "provided"],
    ["tail_uuid", "in", human_uuids]
collection_uuids = map(lambda l: l['head_uuid'], provenance_links['items'])

# build map of human uuid -> PGP ID
pgpid = {}
for pgpid_link in pgpid_links['items']:
  pgpid[pgpid_link['head_uuid']] = pgpid_link['name']

# build map of collection uuid -> PGP ID
for p_link in provenance_links['items']:
  pgpid[p_link['head_uuid']] = pgpid[p_link['tail_uuid']]

# get details (e.g., list of files) of each collection
collections = arvados.api('v1').collections().list(filters=[
    ["uuid", "in", collection_uuids]

# print PGP public profile links with file locators
for c in collections['items']:
  for f in c['files']:
    print "https://my.pgp-hms.org/profile/%s %s %s%s" % (pgpid[c['uuid']], c['uuid'], ('' if f[0] == '.' else f[0]+'/'), f[1])

https://my.pgp-hms.org/profile/hu43860C a58dca7609fa84c8c38a7e926a97b2fc var-GS00253-DNA_A01_200_37-ASM.tsv.bz2
https://my.pgp-hms.org/profile/huB1FD55 ea30eb9e46eedf7f05ed6e348c2baf5d var-GS000010320-ASM.tsv.bz2
https://my.pgp-hms.org/profile/huDF04CC 4ab0df8f22f595d1747a22c476c05873 var-GS000010427-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu7A2F1D 756d0ada29b376140f64e7abfe6aa0e7 var-GS000014566-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu553620 7ed4e425bb1c7cc18387cbd9388181df var-GS000015272-ASM.tsv.bz2
https://my.pgp-hms.org/profile/huD09534 542112e210daff30dd3cfea4801a9f2f var-GS000016374-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu599905 33a9f3842b01ea3fdf27cc582f5ea2af var-GS000016015-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu43860C a58dca7609fa84c8c38a7e926a97b2fc+302 var-GS00253-DNA_A01_200_37-ASM.tsv.bz2
https://my.pgp-hms.org/profile/huB1FD55 ea30eb9e46eedf7f05ed6e348c2baf5d+291 var-GS000010320-ASM.tsv.bz2
https://my.pgp-hms.org/profile/huDF04CC 4ab0df8f22f595d1747a22c476c05873+242 var-GS000010427-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu7A2F1D 756d0ada29b376140f64e7abfe6aa0e7+242 var-GS000014566-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu553620 7ed4e425bb1c7cc18387cbd9388181df+242 var-GS000015272-ASM.tsv.bz2
https://my.pgp-hms.org/profile/huD09534 542112e210daff30dd3cfea4801a9f2f+242 var-GS000016374-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu599905 33a9f3842b01ea3fdf27cc582f5ea2af+242 var-GS000016015-ASM.tsv.bz2
https://my.pgp-hms.org/profile/hu599905 d6e2e57cd60ba5979006d0b03e45e726+81 Witch_results.zip
https://my.pgp-hms.org/profile/hu553620 ea4f2d325592a1272f989d141a917fdd+85 Devenwood_results.zip
https://my.pgp-hms.org/profile/hu7A2F1D 4580f6620bb15b25b18373766e14e4a7+85 Innkeeper_results.zip
https://my.pgp-hms.org/profile/huD09534 fee37be9440b912eb90f5e779f272416+82 Hallet_results.zip

Search for a variant

Now we will use crunch to issue a ‘grep’ job to look for variant rs1126809 in each of the “var-” files (these contain variant calls from WGS data).

>>> job = {}
for c in collections['items']:
  if [] != filter(lambda f: re.search('^var-.*\.tsv\.bz2', f[1]), c['files']):
    job[c['uuid']] = arvados.api('v1').jobs().create(body={
      'script': 'grep',
      'script_parameters': {'input': c['uuid'], 'pattern': "rs1126809\\b"},
      'script_version': 'e7aeb42'
    print "%s %s" % (pgpid[c['uuid']], job[c['uuid']]['uuid'])

hu43860C qr1hi-8i9sb-wbf3uthbhkcy8ji
huB1FD55 qr1hi-8i9sb-scklkiy8dc27dab
huDF04CC qr1hi-8i9sb-pg0w4rfrwfd9srg
hu7A2F1D qr1hi-8i9sb-n7u0u0rj8b47168
hu553620 qr1hi-8i9sb-k7gst7vyhg20pt1
huD09534 qr1hi-8i9sb-4w65pm48123fte5
hu599905 qr1hi-8i9sb-wmwa5b5r3eghnev
hu43860C qr1hi-8i9sb-j1mngmakdh8iv9o
huB1FD55 qr1hi-8i9sb-4j6ehiatcolaoxb
huDF04CC qr1hi-8i9sb-n6lcmcr3lowqr5u
hu7A2F1D qr1hi-8i9sb-0hwsdtojfcxjo40
hu553620 qr1hi-8i9sb-cvvqzqea7jhwb0i
huD09534 qr1hi-8i9sb-d0y0qtzuwzbrjj0
hu599905 qr1hi-8i9sb-i9ec9g8d7rt70xg

Monitor job progress by refreshing the Jobs page in Workbench, or by using the API:

>>> map(lambda j: arvados.api('v1').jobs().get(uuid=j['uuid']).execute()['success'], job.values())
[None, True, None, None, None, None, None, None, None, None, None, None, None, None]

Unfinished jobs will appear as None, failed jobs as False, and completed jobs as True.

After the jobs have completed, check output file sizes.

>>> for collection_uuid in job:
  job_uuid = job[collection_uuid]['uuid']
  job_output = arvados.api('v1').jobs().get(uuid=job_uuid).execute()['output']
  output_files = arvados.api('v1').collections().get(uuid=job_output).execute()['files']
  # Test the output size.  If greater than zero, that means 'grep' found the variant
  if output_files[0][2] > 0:
    print("%s has variant rs1126809" % (pgpid[collection_uuid]))
    print("%s does not have variant rs1126809" % (pgpid[collection_uuid]))

hu553620 does not have variant rs1126809
hu43860C does not have variant rs1126809
hu599905 has variant rs1126809
huD09534 has variant rs1126809
hu553620 does not have variant rs1126809
huB1FD55 does not have variant rs1126809
huDF04CC has variant rs1126809
hu7A2F1D has variant rs1126809
hu7A2F1D has variant rs1126809
hu599905 has variant rs1126809
huDF04CC has variant rs1126809
huB1FD55 does not have variant rs1126809
huD09534 has variant rs1126809
hu43860C does not have variant rs1126809

Thus, of the 14 WGS results available for PGP participants reporting non-melanoma skin cancer, 8 include the rs1126809 variant.

Previous: Scripts provided by Arvados

The content of this documentation is licensed under the Creative Commons Attribution-Share Alike 3.0 United States licence.
Code samples in this documentation are licensed under the Apache License, Version 2.0.