Provenance Design Pattern
Authors and contributors:
- Chris Mungall (author)
Date: 2013
Document Type: metadata_description
In this ontology we attempt as far as possible to record provenance at the per-relationship or per-axiom level, as opposed to at the per-class level. This additional level of granularity allows us to disentangle conflicting statements or add/remove relationships from specific sources.
The corresponding OWL construct that allows us to do this is an axiom annotation.
This document describes the use of axiom annotations in Uberon. The standard is evolving, and we welcome comments or suggestions for improvement.
In OWL, a class axiom is a statement found in an ontology, such as:
Class: 'scleral cartilage'
SubClassOf: develops_from some 'neural crest'
We may wish to record the origin of this assertion (especially in the case of developmental relationships, which straddles the border between ontological knowledge and experimental data). We can do this in OWL by associating the axiom with any number of Annotation - essentially key-value pairs, with the keys coming from some standard vocabulary.
An annotation on an annotation looks like this in Manchester syntax:
Class: 'scleral cartilage'
SubClassOf:
Annotation: source 'PMID:17051547'
develops_from some 'neural crest'
In Protege, the annotation is accessed via the '@' button next to the axiom. This symbol is colored if there exists any axioms for that annotation.
In OWL you can be fairly flexible with your axiom annotations, but certain conventions make interoperability with OBO-Format easier. These include writing publications in the above format (although in future we would prefer to include the full URL of the PMID).
The above looks like this in obo-format:
id: UBERON:0010289
name: scleral cartilage
relationship: develops_from UBERON:0002342 {source='PMID:17051547'} ! neural crest
For historical reasons, Uberon includes many external ontologies as sources of axioms. Typically the ontology name is used, e.g.
source='ZFA'
Note that in these cases the source was a species ontology, but the axiom has been generalized
In all cases, ontology-based axiom annotations should be treated more as an audit trail rather than strong evidence in itself. We would like to arrive at a situation where the provenance for each ontology is some combination of domain expert curator and review, textbook - or in some cases, primary literature.
Note also that currently no attempt has been made to capture all the ontologies that support a particular generalized statement. We may do this in the future, but the highest priority is getting good evidence for each assertion that is reasonably applicable for the appropriate taxon.
Ideally axioms should supported by publications. Here the convention is to use a 'source' annotation together with an obo-style identifier such as a PMID or ISBN.
For textbooks that are used frequently (e.g. Kardong in the case of uberon), these are referenced in the ontology header.
Many axioms in Uberon are supported by Wikipedia. Typically this was just noted as source='Wikipedia'. A dated wikipedia page would be better, but in many cases Wikipedia should be seen as a temporary measure, and eventually the provenance would be at the level of the references cited in the wikipedia article.
Often a relationship is supported by a discussion, e.g. on a mail list or tracker. In these cases a URL to the discussion should be provided.
Sometimes batches of relationships are added at meetings. We follow the GO strategy of making a unique identifier for the meeting, providing a stable URL, and using that. Sometimes the meeting will be written up as a design document.
Sometimes the evidence will be an article that is part of this set of articles. In this case, the URL is provided.
For example, the Phenotype RCN neural crest meeting notes has this URL:
In many cases a relationship in an ontology can be automatically inferred. Although we don't typically assert these in the editors versions, these will be visible in the release version with an annotation of is_inferred=true (this is placed automatically by Obo Ontology Release Tool, OORT).
Sometimes this tag will be supplemented by additional URLs. E.g.
Uberon makes use of non-standard inference procedures that are made using a rule-based system outside what normal OWL reasoners can handle. These should be marked with a URL describing them. For example, the axiom in Uberon:
'metaphysis of femur' SubClassOf part_of some 'diaphysis of femur'
Is accompanied by an axiom annotation pointing to Inferring-part-of-relationships
As well as providing a path back to a web document or piece of literature, axioms may have individual notes from curators. The comment property can be used here.
Sometimes semi-structured notes are added by using ad-hoc properties. For example, the axiom that states the stapes develops from Reciherts cartilage has two annotations:
- source: 'classical definition'
- contradicted_by: doi:10.1111/j.1469-7580.2006.00524.x
This is somewhat ad-hoc. Conventions are still evolving and these annotations may be restructured to fit into a more standard community agreed upon model later on.
Our goal is to eventually have each assertion backed up using an evidence system as in GO. Currently the focus is on capturing the reference, although we may go back later and add evidence types/codes.
This article describes how to record provenance about a relationship, not when and whether a curator should make a relationship. There are separate articles for this. See for example Evolutionary-variability-GCIs - which describes what to do for relationships that only hold for a subset of taxa.
Uberon is a multi-species anatomy ontology and knowledge base, find out more on the home page