=Paper=
{{Paper
|id=Vol-3939/short1
|storemode=property
|title=(Re-)bridging the Anatomy Ontologies with SSSOM
|pdfUrl=https://ceur-ws.org/Vol-3939/short1.pdf
|volume=Vol-3939
|authors=Damien Goutte-Gattat
|dblpUrl=https://dblp.org/rec/conf/icbo/Goutte-Gattat24
}}
==(Re-)bridging the Anatomy Ontologies with SSSOM==
https://ceur-ws.org/Vol-3939/short1.pdf
(Re-)bridging the anatomy ontologies with SSSOM
Damien Goutte-Gattat1,∗ , Nicolas Matentzoglu2
1
University of Cambridge, Downing Street, Cambridge, CB2 3DY, United Kingdom
2
Semanticly, Athens 10563, Greece
Abstract
Ontologies that describe the anatomical structures and cell types of model organisms are critically required for
the annotation and successful exploitation of high-throughput datasets. The Uberon ontology (for anatomical
structures) and the Cell Ontology (CL, for cell types) jointly aim to provide a consistent ontology that can be used
across a wide range of species, by leveraging existing species-centric anatomy ontologies to construct a single
integrated multi-species ontology. In this paper, we describe how we overhauled the integration mechanism
between Uberon, CL, and the species-centric ontologies by using the newly devised SSSOM (“Simple Standard for
Sharing Ontological Mappings”) standard to manage the mappings between all concerned ontologies.
Keywords
Anatomy ontologies, mappings, cross-species studies
1. Introduction
One of the prerequisites for the reuse and reanalysis of datasets from high-throughput experimental
methods – such as single-cell RNA sequencing datasets – beyond their lab of origin is the annotation
of samples and results with standardised terms. That is, we need controlled vocabularies to properly
record information such as the exact techniques used, the organs or tissues from which the experimental
samples were taken, or the cell types that were identified. Several ontologies have been developed
for this kind of purpose, such as the Experimental Factor Ontology (EFO) to annotate experimental
methods [1], the Drosophila Anatomy Ontology (FBbt) to annotate anatomical structures and cell types
in D. melanogaster [2], the Worm Anatomy Ontology (WBbt) to do the same in C. elegans [3], etc. Most
of those ontologies are grouped under the umbrella of the OBO Foundry [4].
However, comparing datasets across species additionally requires that the ontology terms used to
annotate the datasets are themselves comparable. For example, comparing a dataset of mouse blood
cells with a dataset of fly hemocytes requires that some link (relationship) exist between the term
representing the concept of blood in mice and the term representing the concept of hemolymph in flies.
Ideally, this would in turn require that a single cross-species ontology be used for all datasets, containing
all the terms needed to represent all anatomical structures and cell types with their species-specific
variations, organised in a consistent hierarchy.
2. The Uberon strategy for a multi-species ontology
Building a single cross-species anatomy ontology from the ground up, with terms suitable for all the
model organisms, would be a gigantic task, requiring lots of species-specific expertise that would be hard
to gather in a single project. Instead, a more practical approach was adopted by the Uberon project [5]
which, rather than attempting to describe the anatomy of every model organism, leverages the existing
species-specific anatomy ontologies. Uberon aims to provide a “core” anatomy ontology that describes
anatomical structures in a species-neutral way, along with “bridges” that allow to integrate the more
precise terms from the species-specific ontologies (Figure 1). The result of merging the Uberon core
15th International Conference on Biological and Biomedical Ontology, July 17–19 2024, Enschede, The Netherlands
∗
Corresponding author.
Envelope-Open dpg44@cam.ac.uk (D. Goutte-Gattat); nicolas.matentzoglu@gmail.com (N. Matentzoglu)
Orcid 0000-0002-6095-8718 (D. Goutte-Gattat); 0000-0002-7356-1779 (N. Matentzoglu)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
Figure 1: Uberon and the Drosophila Anatomy Ontology (FBbt) maintain their own hierarchy of terms separately.
Simply merging the two ontologies together (A) would result in a hierarchy that is apparently unified but is, in
effect, split in two independent branches that are solely connected by BFO’s ‘independent continuant’ root term,
preventing any meaningful use. (B) Bridging axioms (dotted red arrows) between Uberon and FBbt allow to
connect Drosophila-specific terms to their taxon-neutral counterparts, thereby crossing the gap between the two
branches.
ontology with the species-specific ontologies along with the corresponding bridges is a product called
composite-metazoan (hereafter CM), a single, consistent cross-species ontology. The Cell Ontology (CL),
an ontology of cell types, follows the same approach [6], and is automatically included in Uberon’s CM
product.
To implement this strategy, Uberon maintains a set of cross-ontology mappings, which are used
to determine where the terms from the species-specific ontologies should be placed in the overall
Uberon hierarchy. Across ontologies of the OBO Foundry, the typical method to represent and maintain
mappings is to use cross-references (or, informally, “xrefs”). It’s a very simple method where, to map a
term 𝑇𝐴 in an ontology A to a term 𝑇𝐵 in a foreign ontology B, 𝑇𝐴 is annotated with a oboInOwl:hasDbXref
property, whose value is the short identifier (“CURIE”) of 𝑇𝐵 . As part of the Uberon release pipeline,
a Perl script automatically extracts those cross-references and generates the bridge files needed to
integrate the species-specific ontologies.
While simple, the cross-reference method has several limitations. (i) It does not allow to record any
metadata about the mapping, such as: who asserted that the two terms should be mapped? On what
basis? When was the mapping reviewed? (ii) The method does not allow for any nuance: either two
terms are mapped or they are not, without room to express more subtle relations. (iii) Cross-references
are embedded within the ontology itself, and are therefore not easily reusable by third parties. (iv)
Cross-references have no clear and universally agreed upon meaning. When two terms are mapped to
each other, the meaning of the mapping is left unspecified, and typically has to be inferred from the
ontologies that are being mapped (for example, a mapping between the species-neutral Uberon and the
Drosophila-specific FBbt can be inferred to be a cross-species mapping). This is made worse by the fact
that cross-references in OBO ontologies are used for many different things beyond just mappings; for
example, when used to annotate a term definition, cross-references typically attribute a source for the
definition, akin to a citation in an academic paper.
3. New approach to cross-ontology mappings
We have devised a new approach to maintain and use cross-ontology mappings in Uberon, centred
around three axes: (i) the use of a new format to represent the mappings, (ii) the creation of dedicated
mapping relations, and (iii) the development of new tools to manipulate the mappings and, in particular,
derive OWL axioms from them.
3.1. The SSSOM standard
The Simple Standard for Sharing Ontological Mappings (SSSOM) is a recently devised standard specifi-
cally intended to facilitate the exchange of semantic mappings [7]. It allows mappings to be treated as
first-class data entities and to attach to them a range of metadata such as provenance and licensing
information. The standard defines a common data model to represent mappings as well as two dis-
tinct serialisation formats to store and transport instances of the model: a JSON-based format and a
TSV-based format. The TSV format has been specifically designed to be easily manipulatable both by
standard spreadsheet software (for editing by curators) and by common science libraries and tools (for
use by data scientists and engineers). At least two independent implementations of the standard are
available: SSSOM-Py and SSSOM-Java, for the Python and Java programming languages, respectively.
In addition, several ontology-related tools have started to add direct support for the format, such as the
Ontology Access Kit [8] and the Ontology Development Kit [9]. The community resource Biomappings
also provides its mappings in SSSOM, among other formats [10].
3.2. New mapping relations for cross-species mappings
One of the benefits brought by the SSSOM standard is the possibility to use highly specific mapping
predicates to express precisely the intended meaning of a mapping. While many mapping sets don’t
actually need this, and can use common mapping predicates such as those from the SKOS vocabulary [11]
(skos:exactMatch, skos:narrowMatch, etc.), cross-species mappings are an example of an application
where the common predicates are not sufficient. For example, let us consider a mapping between the
FBbt term for neuron (FBbt:00005106, representing a neuron in a fruit fly) and the corresponding term
in the Cell Ontology (CL:0000540, representing a neuron in any species): clearly it would not be correct
to state that a fly neuron is the same concept as a neuron in any other species – the two terms are
not interchangeable, so skos:exactMatch is not a suitable mapping predicate. A slightly more correct
predicate could be skos:narrowMatch, because is it true that a fly neuron is a narrower concept than a
species-neutral neuron; but a sensory neuron (CL:0000101) is also a narrower concept than a generic
neuron, yet the relation between ‘sensory neuron’ and ‘neuron’ is of a different nature than the relation
between ‘fly neuron’ and ‘(species-neutral) neuron’.
Overall we believe that cross-species mappings are sui generis mappings and that they warrant the
use of dedicated mapping predicates to reflect that fact. We therefore expanded the SEMAPV vocabulary,
specifically established for use with SSSOM [12], with four new mapping predicates (Figure 2) that
mirror existing SKOS predicates but are specifically intended for cases where the subject and object of a
mapping belong to different taxonomic groupings.
3.3. Deriving OWL bridges from SSSOM
Even as SSSOM adoption spreads rapidly, it is likely that many tools will remain unable to directly
exploit SSSOM sets in the foreseeable future. Therefore, when mappings are stored in the SSSOM format,
we need to be able to convert them into OWL axioms that can be used by any ontology manipulation
tool. To that effect, we have developed a SSSOM plugin for ROBOT, the standard tool used to manipulate
OBO ontologies [13]. The plugin is built on top of the Java implementation of the SSSOM standard and
provides a sssom:inject command that takes a SSSOM mapping set as input, derives OWL axioms from
the mappings, and injects them into an ontology.
Rather than hardcoding the logic for deriving the axioms – which would have been efficient but would
have made the logic harder to modify and re-use – we designed a small domain-specific language (DSL),
named SSSOM/T-OWL (“SSSOM/Transform-to-OWL”), that allows users to describe which mappings
in a set should be transformed into OWL axioms and what kind of axioms should be produced.
Figure 2: Excerpt of the semantic mapping vocabulary (SEMAPV), with the new mapping relations (yellow)
introduced to specifically represent cross-species mappings.
Table 1
SSSOM/T-OWL filters, preprocessors, and generators
SSSOM/T-OWL element Examples
Atomic filter predicate==skos:exactMatch
Negated filter !subject==UBERON:*
Intersection filter subject==UBERON:* && predicate==skos:exactMatch
Union filter subject==UBERON:* || subject==CL:*
Inversion preprocessor -> invert()
Exclusion preprocessor -> stop()
Assignation preprocessor -> assign("subject_source", "uberon.owl")
Replacement preprocessor -> replace("mapping_tool", "pattern", "replacement")
Generic axiom generator -> create_axiom("%subject_id EquivalentTo: %object_id")
Subject annotation generator -> annotate_subject(oboInOwl:hasDbXref, "%object_curie")
Object annotation generator -> annotate_object(oboInOwl:hasDbXref, "%subject_curie")
The SSSOM/T-OWL language is organised around the single concept of “transformation rules”. A
rule is made of two elements (Table 1): a filter (or selector) to decide whether the rule should apply to a
given mapping; and either a preprocessor, to modify a mapping on the fly, or a generator, to produce a
OWL axiom from the mapping.
Filters allow the value of any of the SSSOM metadata slots to be compared with a target value and to
discard any mapping with a different value. A basic form of wildcard matching is supported, where a
filter can select mappings with a value that starts with the same prefix as the target value. Individual
filters can be combined to select mappings based on more than just one metadata slot; they can also be
negated, so that a filter that would select a given set of mappings would, after negation, instead select
the complementary set of mappings.
Preprocessors allow a mapping to be modified before it is used by subsequent rules. Possible
modifications include inverting a mapping (the subject becomes the object and vice-versa) and changing
some of the mapping’s metadata. A preprocessor can also be used to completely remove a mapping
from the set, so that no further rule will be applied to this mapping.
Finally, the generators take the selected mapping and turn it into an axiom. The most important
generator function is create_axiom, which can produce an arbitrary axiom described by an expression
in Manchester syntax [14]. Other generators allow classes of the ontology to be annotated with any of
the available mappings metadata.
Of note, the current implementation of the SSSOM/T-OWL language cleanly separates the OWL
generators from the filters and the preprocessing functions. This makes it easy to create other domain-
Figure 3: Uberon’s new pipeline. A Single SSSOM set, encompassing mappings with all target species, is created
by extracting old-style cross-references from both Uberon and several other species-specific anatomy ontologies
– except FlyBase’s FBbt, which already provides a ready-to-use SSSOM set (1). The resulting set is processed by
the Uberon SSSOM/T-OWL ruleset (2) to create the bridge files, which are then merged along with the source
ontologies to create the collected-metazoan ontology (3). The final composite-metazoan is created (4) by pruning
redundant terms using a custom ROBOT plugin.
specific languages similar to SSSOM/T-OWL, using the same overall syntax, the same filters and the
same preprocessors, but that generate objects other than OWL axioms.
4. Application of the approach to the Uberon pipeline
With the components described in the previous section in place, we overhauled the pipeline that builds
Uberon’s composite-metazoan product (Figure 3).
4.1. Collecting the mappings
Rather than migrating all of Uberon mappings to SSSOM at once, we have adopted a phased approach,
in which we have used the mappings between Uberon and FBbt as a test bed in a preliminary phase.
Uberon/FBbt mappings were initially maintained as cross-references within Uberon, as for all the
other Uberon mappings. We migrated them to a SSSOM mapping set that we now maintain on the
FBbt side: existing FBbt cross-references in Uberon were extracted from the ontology and converted
into SSSOM mappings using the newly minted semapv:crossSpeciesExactMatch predicate (all existing
cross-references were for exact mappings, since that was the only kind of mapping allowed by the
cross-reference system), that are now published alongside the FBbt ontology itself.
Mappings between Uberon and the other species-specific anatomy ontologies beyond FBbt are still
currently maintained as cross-references either within Uberon or within the species-specific ontologies.
They are extracted and converted to SSSOM dynamically as part of Uberon’s release pipeline. In a
second phase, which we have not started yet, we will migrate those mappings fully to SSSOM, as was
done for the FBbt mappings.
4.2. Applying the transformation rules
Regardless of their origin, all mappings between Uberon, CL, and the species-specific ontologies are
processed by a single large SSSOM/T-OWL ruleset that contains all the required logic to generate the
bridging axioms. A typical rule in this ruleset is the following:
( s u b j e c t ==UBERON : ∗ | | s u b j e c t ==CL : ∗ ) && o b j e c t == F B b t : ∗
&& p r e d i c a t e == semapv : c r o s s S p e c i e s E x a c t M a t c h
−> c r e a t e _ a x i o m ( ’ % o b j e c t _ i d E q u i v a l e n t T o : % s u b j e c t _ i d and
( BFO : 0 0 0 0 0 5 0 some NCBITaxon : 7 2 2 7 ) ’ ) ;
That rule can be literally understood as: a mapping between a Uberon (or CL) term and a FBbt term
and that uses the semapv:crossSpeciesExactMatch predicate must be transformed into an equivalence
axiom that states that the FBbt term is equivalent to the intersection of the Uberon (or CL) term and an
existential restriction over BFO:0000050 (‘part of’) and NCBITaxon:7227 (‘Drosophila melanogaster’).
All axioms produced by the SSSOM/T-OWL ruleset are saved into a separate bridge file for each
foreign ontology.
4.3. Building the composite-metazoan ontology
Once the bridge files have been generated, an intermediate ontology (collected-metazoan) is built
by merging together Uberon itself, the Cell Ontology, all the species-specific ontologies, and their
corresponding bridges. Finally, composite-metazoan itself is derived from this intermediate product by
applying some custom logic (implemented in a small, dedicated ROBOT plugin) to “prune” redundant
terms and replace them with equivalent anonymous class expressions.
The details of this last operation are beyond the scope of this paper, but as an example, let us consider
the FBbt term ‘ovary’ (FBbt:00004865): it is mapped to the Uberon term ‘ovary’ (UBERON:0000992), so
collected-metazoan contains the following axiom:
F B b t : 0 0 0 0 4 8 6 5 E q u i v a l e n t T o : UBERON : 0 0 0 0 9 9 2 and
( BFO : 0 0 0 0 0 5 0 some NCBITaxon : 7 2 2 7 )
The “pruning” operation leading to composite-metazoan consists of replacing all occurrences of
FBbt:00004865 by the anonymous expression it is equivalent to. So the following axiom, which states
that the ‘oviduct’ (FBbt:00004911) is ‘continuous with’ (RO:0002150) the fly ovary:
F B b t : 0 0 0 0 4 9 1 1 S u b C l a s s O f : RO : 0 0 0 2 1 5 0 some F B b t : 0 0 0 0 0 4 8 6 5
gets rewritten as:
F B b t : 0 0 0 0 4 9 1 1 S u b C l a s s O f : RO : 0 0 0 2 1 5 0 some ( UBERON : 0 0 0 0 9 9 2 and
( BFO : 0 0 0 0 0 5 0 some NCBITaxon : 7 2 2 7 ) )
5. Conclusion
In this project, we have (i) expanded the SEMAPV vocabulary so that cross-species mappings can be
specifically represented in SSSOM; (ii) introduced a domain-specific language (SSSOM/T-OWL) to allow
converting SSSOM mappings into arbitrary OWL axioms; (iii) implemented said language in a new
ROBOT plugin to inject SSSOM-derived axioms into a OWL ontology; (iv) used those new tools to
update the integration mechanism between Uberon, CL, and the species-centric ontologies, with an
approach centered on the use of SSSOM, rather than cross-references, to manage the mappings between
the ontologies.
As a result, cross-species mappings across anatomy ontologies are now available as a bona fide release
artefact of Uberon, provided as a distinct file in a known location in a standard format, thereby greatly
facilitating the reuse of said mappings by any interested third party. The logic for deriving the bridge
files (needed to correctly merge Uberon and the species-specific ontologies), now expressed in the
SSSOM/T-OWL language, is itself consequently more easily maintainable and reusable.
While several aspects of this project (most notably the precise SSSOM/T-OWL rules we use) are
specific to the needs of Uberon, we believe the general approach of maintaining mappings as SSSOM
sets and transforming them into OWL bridges using a simple DSL could be generalised to any project
that requires integrating several ontologies together.
Acknowledgments
This work was supported by grant BB/T014008 from the UK Biotechnology and Biological Sciences
Research Council (BBSRC) and the US National Science Foundation Directorate of Biological Sciences
(NSF/BIO).
References
[1] J. Malone, E. Holloway, T. Adamusiak, M. Kapushesky, et al., Modeling sample variables with an Ex-
perimental Factor Ontology, Bioinformatics 26 (2010) 1112–1118. doi:10.1093/bioinformatics/
btq099 .
[2] M. Costa, S. Reeve, G. Grumbling, D. Osumi-Sutherland, The Drosophila anatomy ontology,
Journal of Biomedical Semantics 4 (2013) 32. doi:10.1186/2041- 1480- 4- 32 .
[3] R. Y. N. Lee, P. W. Sternberg, Building a Cell and Anatomy Ontology of Caenorhabditis Elegans,
Comparative and Functional Genomics 4 (2003) 121–126. doi:10.1002/cfg.248 .
[4] R. Jackson, N. Matentzoglu, J. A. Overton, R. Vita, et al., OBO Foundry in 2021: Operationalizing
open data principles to evaluate ontologies, Database 2021 (2021) baab069. doi:10.1093/database/
baab069 .
[5] C. J. Mungall, C. Torniai, G. V. Gkoutos, S. E. Lewis, M. A. Haendel, Uberon, an integrative
multi-species anatomy ontology, Genome Biology 13 (2012) R5. doi:10.1186/gb- 2012- 13- 1- r5 .
[6] A. D. Diehl, T. F. Meehan, Y. M. Bradford, M. H. Brush, et al., The Cell Ontology 2016: Enhanced
content, modularization, and ontology interoperability, Journal of Biomedical Semantics 7 (2016)
44. doi:10.1186/s13326- 016- 0088- 7 .
[7] N. Matentzoglu, J. P. Balhoff, S. M. Bello, C. Bizon, et al., A Simple Standard for Sharing Ontological
Mappings (SSSOM), Database 2022 (2022) baac035. doi:10.1093/database/baac035 .
[8] C. Mungall, Harshad, P. Kalita, C. T. Hoyt, et al., INCATools/ontology-access-kit: V0.5.24, Zenodo,
2023. doi:10.5281/zenodo.10277632 .
[9] N. Matentzoglu, D. Goutte-Gattat, S. Z. K. Tan, J. P. Balhoff, et al., Ontology Development Kit: A
toolkit for building, maintaining and standardizing biomedical ontologies, Database 2022 (2022)
baac087. doi:10.1093/database/baac087 .
[10] C. T. Hoyt, A. L. Hoyt, B. M. Gyori, Prediction and curation of missing biomedical identifier
mappings with Biomappings, Bioinformatics 39 (2023) btad130. doi:10.1093/bioinformatics/
btad130 .
[11] S. Bechhofer, A. Miles, SKOS Simple Knowledge Organization System Reference, W3C Recommen-
dation, W3C, 2009.
[12] N. Matentzoglu, J. Flack, J. Graybeal, N. L. Harris, et al., A Simple Standard for Ontological
Mappings 2022: Updates of data model and outlook, in: 17th International Workshop on Ontology
Matching, volume 3324, CEUR-WS.org, Hangzhou, China, 2022, pp. 61–66.
[13] R. C. Jackson, J. P. Balhoff, E. Douglass, N. L. Harris, C. J. Mungall, J. A. Overton, ROBOT:
A Tool for Automating Ontology Workflows, BMC Bioinformatics 20 (2019). doi:10.1186/
s12859- 019- 3002- 3 .
[14] P. Patel-Schneider, M. Horridge, OWL 2 Web Ontology Language Manchester Syntax (Second
Edition), W3C Note, W3C, 2012.
A. Online Resources
The SSSOM-Java project, which implements both the SSSOM standard and the SSSOM/T-OWL language
described in this paper, is hosted on GitHub.
Uberon artefacts concerned by this project are available under the following permanent URLs:
Artefact PURL
SSSOM mapping set http://purl.obolibrary.org/obo/uberon/uberon.sssom.tsv
Bridge to ontology X http://purl.obolibrary.org/obo/uberon/bridge/uberon-bridge-to-X.owl
Composite Metazoan http://purl.obolibrary.org/obo/uberon/composite-metazoan.owl