A Simple Standard for Sharing Ontology Mappings (SSSOM)

Matentzoglu, Nicolas; Balhoff, James P.; Callahan, Tiffany; Chute, Christopher; Dahzi, Jiao; Duncan, William; Gabriel, Davera; Graybeal, John; Haendel, Melissa; Harmse, Henriette; Harold, Solbrig; Harris, Nomi; Hegde, Harshad; Hoyt, Charles Tapley; Jimenez-Ruiz, Ernesto; Jupp, Simon; Kim, Hyeongsik; Koehler, Sebastian; Liener, Thomas; Malone, James; McLaughlin, James; Munoz-Torres, Monica; Osumi-Sutherland, David; Overton, James; Thessen, Anne; Vasilevsky, Nicole; Mungall, Chris

​

WSBO - 14.07.2021

​

https://fairsharing.org/bsg-s001618

​

A plaidoyer for caring more about the standardisation and dissemination

of ontology mappings.

What is a mapping (in the sense of this talk)?

• a mapping is a correspondence of two terms
• a mapping set is a set of term mappings, usually assembled for a particular purpose
• an alignment is a special kind of mapping set which contains all mappings between two ontologies

​

​

MONDO:0006071

“adenofibroma”

MONDO:0006071

EFO:1000070

UBERON:0000483

"epithelium (drosophila)" �FBbt:00007005

NCBITaxon:7227

Type 1: string - term

Type 2: term - term

Type 3: complex

Convergence vs Mapping

• Mapping and Convergence are two complementary approaches to achieve synergy
• Convergence: continuous ontological unification (OBO Foundry)
• Mappings: bridging across ontologies (UMLS, OxO)
• Convergence is the ideal, but the idea of total convergence is unrealistic
• Extremely expensive, hard to agree on, “political” issues
• In Medical Terminology, it is not even considered anymore
• Mapping is the reality, and they are are essential for bridging semantic spaces/silos

How are mappings curated and used in practice?

• Monarch Initiative, IMPC and MOD databases: Cross-species mappings
• Increasing the clinical relevance of model organism data: search, variant prioritisation
• Using an automated logic-based approach to identify similar phenotypes
• IHCC: Curators map data dictionaries for description of rare disease cohorts to an application ontology for genomics cohorts
• Bosch: Enriching Bosch's internal ontologies with external ones such as FoodOn, Wikidata, and GAO, etc. (to support smart assistant agents and recommendation platforms)

Mappings are essential to bridge semantic spaces, but they are hard to use.. and share.

Non-transparent imprecision: Mappings are rarely exact equivalents.. but we often don’t know that

• Precision: exact, narrow, broad, close, related
• Imprecise is not necessarily bad...
• ... but non-transparent imprecision is
• Aggravated when considering “crosswalks”
• inverse walking (Alzheimer’s → Alzheimer’s 2):
• multihop walking (Alzheimer’s 2 → Alzheimer’s → Alzheimer’s 3)
• hairball problem
• You cannot use bidirectional nor multi-hop crosswalks without knowing the precision

O2:Alzheimer’s

O1:Alzheimer’s 2

O3:Alzheimer’s 3

(broad)

(narrow)

Inaccurate and incomplete: The consequence of relying on tools or mass manual review

• New tooling make large scale matching more viable - but precision and accuracy will continue to be questionable
• Curators are basically insanely accurate at this - but for 20K mappings?
• The moment a mapping is “complete”, it is basically out of date.
• (Aside: we keep on redoing the same mappings over and over again - which does not help “mapping” convergence and costs so many hours)

An overview of mapping systems in the biomedical domain (excerpt)

• Alignment API and EDOAL: Expressive and Declarative Ontology Alignment Language
• Bioportal Mappings
• OBO hasDbXref (-> OxO)
• Others:
• SNOMED simple and complex maps, Common Terminology Services 2
• OpenPHACTS Linksets, BridgeDB Mapping Vocabulary
• SEMAF: Flexible Semantic Mapping Framework (emerging)
• SEMAF registries / Mapping Commons
• Formalising crosswalks great idea
• Biohackathon 2015: Ontology Mapping Metadata

Mappings are pretty unFAIR

• Findable:
• Publish {mappings} in dedicated repositories
• Minimum {mappings} metadata
• Use unique Ids to identify concepts, relations and {mapping sets}
• Accessible:
• Use simple APIs to provide access to {ontology} content and metadata (Bioportal, OLS)
• Interoperable:
• Use common practices at syntactic level
• Use common structure for {concepts and relations}
• Reusable:
• Reuse existing related {mappings}
• Provide rich metadata to describe {mappings}
• without knowing the curation rules, many high �precision scenarios would not even permit reuse

prov:stolen_from

_:b

Mappings could be:�

• transparently inaccurate
• transparently incomplete
• transparently imprecise
• transparently conflicting
• open
• FAIR and easy to use

​

​

7

Making mappings FAIR and Open: A Simple Standard for Sharing Ontology Mappings (SSSOM)

mapping / correspondence

mapping set

mapping server

mapping commons

derives

makes available & trusts

makes available & trusts

maintains

part of

derived �mapping set

partial �mapping set

alignment

https://fairsharing.org/bsg-s001618

Mapping model

match_type

HumanCurated

confidence

0.9

match_type

HumanCurated

subject_id

EFO:1000070

object_id

MONDO:0006071

predicate_id

skos:exactMatch

match_type

HumanCurated

subject_label

Adenofibroma

object_label

adenofibroma

mapping_date

29.05.2021

orcid:0001-.....

subject_match_field

label

The SSSOM metadata model

Rich YAML schema powered by

Shex shapes for validating rdf

JSON Schema

Markdown docs

- subject_id

- subject_label

- subject_category

- predicate_id

- predicate_label

- object_id

- object_label

- object_category

- match_type

- creator_id

- creator_label

- license

- subject_source

- subject_source_version

- object_source

- object_source_version

- mapping_provider

- mapping_cardinality

- mapping_tool

- mapping_date

- confidence

- subject_match_field

- object_match_field

- match_string

- subject_preprocessing

- object_preprocessing

- match_term_type

- semantic_similarity_score

- see_also

- other

- comment

https://github.com/mapping-commons/sssom-py/tree/master/schema

Example SSSOM tsv file

Metadata header with curie map, and various mapping set level metadata fields.

Actual mappings with metadata in TSV form

pd.read_csv(f,comment=”#”, sep=”\t”)

sssom-py: a python toolkit and CLI to process SSSOM files

obographs-json

alignment-api-xml

alignment-api-xml

import

import

sssom-tsv

sssom-tsv

export

sssom-tsv

merge

merge

filter

https://github.com/mapping-commons/sssom-py

Example uses

• Mouse-Human phenotype Mapping Commons
• Pistoia Alliance contributed a lot of automated mappings
• Mapping across disease and phenotype ontologies (merged into FAIR implementation project)
• Strong focus on automated matching approaches to retain currency
• IMPC example
• Similar efforts for other MODs
• The National Microbiome Data Collaborative (NMDC, https://microbiomedata.org/):
• To provide standardized metadata for data provided by GOLD, map GOLD database fields to MIxS and environmental data elements to ENVO
• Others:
• CCDH (mapping clinical data models and value sets to ontologies)
• biomappings
• omop2obo

A simple guide to make your mappings FAIR.. and open

• Publish mappings:
1. in a standard file format (tsv, csv, json, xml) - not (only) buried in a database
2. using a standard open license
3. referring to a standard metadata schema such as SSSOM or EDOAL
4. in a public space with an issue tracker (GitHub)
5. with at least the following metadata:
• MUST: precision, confidence, match type, prefix map, license
• SHOULD: curation rules / mapping justification
• (It is hard to stop here as there are about a dozen or so essential things)
• Nico’s personal amendment: Publish mappings
• in CC0 or CC-BY (provenance!)
• as CSV or TSV to enable straight forward integration into standard data science pipelines
• in a designated Mapping Commons which provides exports from your mappings into SSSOM
• with rich metadata, such as versions of the term sources, provenance, mapping tools used
• Treating mappings with the same love and care as our ontologies and vocabularies

Takeaways

• Striving for convergence does not mean we should dismiss mapping - the two are complementary processes and need to evolve side-by-side�
• Your mappings are valuable, sharing them in a standard format on a public repository does them and your hard work justice - and saves other people a ton or work (OAEI & manual curation)�
• When mapping terms, please curate at the very least precision, confidence, prefix maps and curation rules.�
• Get involved in SSSOM and influence its development as a community standard

Join us for the next SSSOM user meetup on 3rd September..

Acknowledgements:

​

Phenomics First (NIH / NHGRI #1RM1HG010860-01): Spec, Mondo integration, sssom-py CLI

​

Bosch Gift to LBNL: sssom-py IO, testing, converters

​

Mapping community: Lots of volunteering contributions, e.g. Charlie Hoyt (sssom-py), Pistoia Alliance (Thomas Liener), John Graybeal (spec), James McLaughlin (infrastructure), ...

​

Core team: https://w3id.org/sssom/SSSOM.md