OntoVoc Report - Sprint 11/2018#
The objectives of this first sprint in November 2018 were to:
define criteria for suitable ontologies
identify ontologies for five fields/keywords of the MiAIRR data standard and
assess technical aspects of ontology integration into databases
The Team initially discussed an approach where only vocabularies (i.e. lists of terms) and not ontologies (i.e. many terms connected by predicates) would have been defined. These vocabularies would have been derived from ontologies, but this process would not necessarily have been reversible. The notion at this time point was, that such an approach would allow to solve a number of problems like combining multiple sources and removing duplicated leaves. However, after some discussions this approach was effectively abandoned for a number of reasons:
It would discard the UID for an entity. As the UID (in contrast to the name string) is guaranteed to be stable and unique, it facilitates updates, linking and information representation, all of which would otherwise be lost.
In general, it will be more sustainable to work with the maintainers of an existing ontology to include entities/terms, than just dumping their terms into a list and adding new ones.
Well-designed ontologies will not contain duplicated entities, although they might appear to do so in a simple browsers (i.e. this is an artifact of representation). Ontologies that actually do contain duplicates are excluded by criterium 2.
Criteria for Ontologies#
Ontologies used within AIRR standards
MUST  cover the majority of the required terms, but complete coverage is OPTIONAL
MUST have a structure that is scientifically correct and logically coherent
MUST NOT feature complexity that makes it hard to use for queries and data representation
SHOULD already be widely adopted
MUST be actively maintained
MUST be available under a free license
Comments on criteria:
ad (1): For most fields it will be difficult to find complete and accurate ontologies. Therefore picking the best available ontology and working with its maintainers to include missing terms is expected to be the most sustainable approach.
ad (6): A number of ontologies need to be licensed from their respective copyright holders. This results in potential barriers for implementation and distribution of such ontologies. Therefore only ontologies available under a free license are considered suitable for AIRR-compliant databases. The list of suitable licenses is not final, but includes: CC0 and CC BY.
(designations are MiAIRR field names and
Cell subset (
multicellular anatomical structure
/BFO_0000002/BFO_0000004/anatomical entity/material anatomical entity/anatomical structure/multicellular anatomical structure
License: CC BY
latest release: 2018-10-15
Maintainer: Chris Mungall, LBL, CA, US (email@example.com)
Strain name (
one ontology is not enough
UID assigned by ontologies are guaranteed to be unique and permanent .
A repository MAY use internal identifiers that are distinct from UIDs. However, to be AIRR-compliant it MUST be able to map UIDs to its identifiers.
Points of “AIRR compliance” would typically be:
When data is extracted from the repository through a Query API (CRWG)
When data is extracted from the repository into a file format (DataRep)
Integration of ontologies into repositories:
There are two main ontology providers offering a REST API and all the ontologies listed above:
NCBO can apparently be slow and sometimes not that stable, while OLS seems to be more stable and potentially has a better long-term support.
Remote ontology services tend to be slow and create external dependencies. On the other hand, while local hosting of an ontology is possible (and partially supported by NCBO and OLS), it requires non-negligible resources. The Team’s current assumption is that queries to remote ontology services can be substantially accelerated if only the relevant section of a respective ontology is queried. Therefore a local service would not be necessary.
Repositories should store both the IDs and the values in their database. This way, they do not have to query the ontology in a scenario where human-readable output is required. In the case of changes, most ontologies try to follow the practice of not changing a term value but instead create a new term with the new value and a new ID, and deprecating the old term. Therefore term deprecation needs to be handled by the repository.
Like for the databases, also the API should be able to handle both IDs and values as query input and return both during a query.
Note that the complete IEDB can be downloaded as SQL dump, it is licensed under CC BY. At a first glance, the main overlap seems to be with
strain_nameand to a smaller extent
disease_diagnosis. However, sample information like
tissueseems to be largely absent from IEDB, so it could currently not be the one-stop solution for AIRR.
MUST / REQUIRED: Indicates that an element or action is necessary to conform to the standard.
SHOULD / RECOMMENDED: Indicates that an element or action is considered to be best practice by AIRR, but not necessary to conform to the standard.
MAY / OPTIONAL: Indicates that it is at the discretion of the user to use an element or perform an action.
MUST NOT / FORBIDDEN: Indicates that an element or action will be in conflict with the standard.