Aber-OWL: Ontology-Based Data Access
- Aber-OWL is a framework for ontology-based data access in biology that automatically infers relationships using formal axioms from OWL-compatible ontologies.
- It integrates a repository, reasoning services, and web interfaces to transform Manchester OWL queries into expanded semantic searches across literature and Linked Data.
- By employing ELK for preclassification and trie-based indexing for efficient search, Aber-OWL enhances precision and scalability in biomedical semantic retrieval.
Searching arXiv for the specified paper to ground the article with the cited source. Aber-OWL is a framework for ontology-based data access in biology that provides reasoning services for bio-ontologies expressed in OWL or OWL-compatible formats such as OBO. It combines an ontology repository, web services, and web interfaces to enable semantic access to biological data and literature through automated reasoning over formal axioms rather than over labels and asserted hierarchies alone. Its central design objective is to expose the “semantic content” of ontologies by computing logical consequences and using the resulting inferred class sets to drive retrieval over literature corpora and Linked Data resources (Hoehndorf et al., 2014).
1. Motivation and conceptual basis
Biological and biomedical resources increasingly use ontologies to annotate data, and these ontologies increasingly contain large volumes of formalized knowledge commonly expressed in the Web Ontology Language. In this setting, the informative content of an ontology is not limited to labels, metadata, or asserted subclass relations; it also includes formal axioms from which additional facts and relationships can be derived. Automated reasoning is therefore required to access the semantic content of ontologies.
The framework is motivated by limitations in portals such as BioPortal, OLS, and Ontobee. These systems provide browsing, metadata, asserted hierarchy, and REST APIs, but they do not expose inferences. Without reasoning, incoherences or unsatisfiable classes cannot be detected, incomplete data cannot be enriched with inferred concepts, queries using higher-level concepts cannot uniformly retrieve items annotated with more specific inferred subclasses, and heterogeneous datasets cannot be uniformly accessed using ontology knowledge as a bridge (Hoehndorf et al., 2014).
Within this perspective, ontology-based data access uses ontologies to specify features of interest at the knowledge level and then uses automated reasoners to obtain all classes satisfying those features. These inferred sets can then be used to retrieve ontology-annotated data or literature. A plausible implication is that Aber-OWL operationalizes OBDA as a service layer for biological ontologies rather than merely as an ontology browser.
2. System architecture and component interaction
Aber-OWL consists of an ontology repository, reasoning services, web services, and web interfaces. The ontology repository is configured with a list of ontology document URIs in the form of OWL files, retrieved via the OWL API. For each ontology and all the ontologies it imports, the system extracts rdfs:label and IAO_0000115 definition annotations. All labels from the ontology and its imports are stored in a trie, or prefix tree, for efficient label search and completion (Hoehndorf et al., 2014).
On service startup, each ontology is classified with ELK under the OWL 2 EL profile. This classification computes the inferred subsumption hierarchy, including the most specific superclasses and subclasses. If a query references an ontology URL that is not already in the repository, the system attempts to fetch, classify, and use it on-the-fly.
The web services accept queries in Manchester OWL Syntax, transform them to OWL class expressions via the OWL API, and run reasoning tasks to return subclasses, superclasses, or equivalent classes. Results are returned in JSON as an array of objects containing the ontology document URI, class IRI, rdfs:label, and IAO_0000115 definition. A JQuery/AJAX web user interface drives these services for interactive query construction and result visualization, and links outward to Aber-OWL: PubMed and Aber-OWL: SPARQL so that a semantic query can be extended into literature or data retrieval workflows (Hoehndorf et al., 2014).
The interaction pattern is direct: the repository and reasoner produce inferred class sets from Manchester queries, web services deliver these sets to clients, and the PubMed and SPARQL modules consume them to expand text or data queries. This suggests a separation between ontology reasoning and downstream retrieval, with the inferred class set serving as the main interoperability object.
3. Ontologies, imports, updates, and deployment
The hosted instance imports the complete library of OBO Foundry ontologies and several user-requested ontologies. Examples named in the description include the Gene Ontology, the Human Phenotype Ontology, and Uberpheno, a cross-species phenotype ontology built from species phenotype ontologies. User-specified ontologies can also be provided by URL (Hoehndorf et al., 2014).
Imports are treated as first-class inputs. For each ontology, Aber-OWL processes labels and definitions not only from the ontology itself but also from all imported ontologies, ensuring that queries can resolve terms defined across import chains. Authoritative OBO URIs are used for ontology class IRIs.
The system reclassifies the included ontologies on startup. When a query references an ontology URL that is not already present in the repository, the ontology is retrieved and classified on demand. The description does not provide an explicit versioning policy or scheduled update cycle beyond these mechanisms. That absence is technically significant: a plausible implication is that reproducibility across time depends on the versions available at startup or through on-demand retrieval rather than on a fixed archival snapshot.
Aber-OWL is freely available at http://aber-owl.net, and the source code is available at https://github.com/reality/SparqOWL. It can be installed locally to provide semantic access to user ontologies and to support semantic queries against local or remote data. Deployment consists of configuring the repository with ontology document URIs, using the OWL API to retrieve ontologies and imports, and selecting a reasoner through the OWL API interface, with ELK as default and HermiT or Pellet as alternatives in custom deployments (Hoehndorf et al., 2014).
4. Reasoning model and query semantics
Aber-OWL uses the OWL 2 EL profile for fast, polynomial-time reasoning suitable for many biomedical ontologies. ELK is the default reasoner because of scalability, speed, and suitability for EL ontologies, but axioms outside EL are ignored by ELK. Ontology classification computes the inferred subclass and superclass structure, and query answering over class expressions returns inferred subclasses, superclasses, or equivalent classes of a query expression. Detection of unsatisfiable classes is also enabled by classification, although it is not described as a dedicated endpoint (Hoehndorf et al., 2014).
Users express queries in Manchester OWL Syntax. Aber-OWL converts the Manchester string into an OWL class expression through the OWL API, using short-form mappings from URIs to labels. The query type may be subclass, superclass, or equivalent-class, with subclass as default. If parsing fails or ontology retrieval fails, the result is an empty result set.
Several query patterns are explicitly described. The existential restriction
is written in Manchester syntax as part_of some nucleus and returns all classes inferred to be parts of the nucleus. A conjunction such as
is written as part_of some 'apoptotic process' and part_of some regulation and returns processes that are parts of apoptosis and are parts of regulation. A simple named-class query such as 'ventricular septal defect' returns all inferred subtypes; in HPO and Uberpheno this includes tetralogy of fallot because the subsumption
is entailed (Hoehndorf et al., 2014).
The framework also supports conjunctive phenotype definitions. The expression
retrieves Tetralogy of Fallot and its specific subtypes when logical axioms in HPO define them as subclasses of this intersection. Query results can also be interpreted as subsumption or equivalence tests: if a superclass query for tetralogy of fallot returns ventricular septal defect, then the former is subsumed by the latter; if an equivalent-class query for a complex expression returns a named class, then that named class is equivalent to the expression. The example of apoptotic [DNA](https://www.emergentmind.com/topics/diverse-negative-attributes-dna-optimization) fragmentation being equivalent to DNA catabolic process and part_of some apoptotic process is presented as illustrative of the pattern, with the caveat that exact examples depend on the ontology’s axioms.
5. Literature and Linked Data access
A major feature of Aber-OWL is the reuse of inferred class sets for literature and data retrieval. Aber-OWL: PubMed implements a four-step workflow. First, a Manchester query is run to obtain all named classes satisfying the expression. Second, a Lucene OR query is constructed from the rdfs:label values of all returned classes. Third, this query is run against MEDLINE/PubMed titles and abstracts and PubMed Central full texts indexed from 2014 corpora using Lucene with the English analyzer, lowercasing, and stopword removal. Fourth, articles mentioning any of the labels are returned, with highlighted spans where the labels occur (Hoehndorf et al., 2014).
The semantic effect is that querying for ventricular septal defect can retrieve articles mentioning tetralogy of fallot, because Tetralogy of Fallot is inferred as a subclass of ventricular septal defect in HPO or Uberpheno. Conjunctive literature queries are supported by running multiple Aber-OWL queries and combining them conjunctively in Lucene. The system therefore uses ontology reasoning to expand textual retrieval beyond direct lexical mention of the original query term.
Aber-OWL: SPARQL extends SPARQL with an OWL block that embeds Aber-OWL results directly into a query over ontology-annotated data. The block syntax is:
1 |
OWL [querytype] [<Aber-OWL service URI>] [<ontology URI>] { [Manchester OWL query] } |
Two embedding forms are supported. In a VALUES binding, the returned class IRIs are bound to a variable for annotation matching. In FILTER IN, results are restricted to those whose annotation IRIs are in the returned set. Prefix or URI rewriting supports translation from OBO-style IRIs such as http://purl.obolibrary.org/obo/GO_0008150 to endpoint-specific schemes such as http://purl.uniprot.org/go/0008150 via prefix forms like GO:0008150 (Hoehndorf et al., 2014).
Two use cases are described. In UniProt, the query part_of some 'apoptotic process' over GO yields all GO terms satisfying the corresponding existential restriction, which are then bound to ?ontid and matched via up:classifiedWith to retrieve all human proteins annotated to part-of-apoptosis terms. In GWAS Central, a subclass query for ventricular septal defect over HPO yields all HPO classes subsumed by that phenotype, and the SPARQL query filters annotations to those inferred classes while combining the result with a significance condition such as p-value ≤ 1e-10. The general pattern is that reasoning is performed outside the endpoint, while the endpoint processes asserted triples during SPARQL evaluation.
6. Interfaces, APIs, optimization strategies, and constraints
The core service is available at http://aber-owl.net/aber-owl/service/. Conceptually, it accepts a query parameter containing a Manchester OWL class expression, a querytype parameter taking subclass, superclass, or equivalent with subclass as default, and an optional ontology parameter specifying an ontology document URI. If the ontology parameter is omitted, the query is evaluated against all ontologies in the repository. The JSON output contains ontologyDocumentURI, classIRI, rdfsLabel, and iaoDefinition when available. Invalid syntax or unavailable ontology returns an empty result set (Hoehndorf et al., 2014).
The PubMed interface is available at http://aber-owl.net/aber-owl/pubmed/. Aber-OWL: SPARQL provides an expansion service that replaces OWL blocks with lists of class IRIs using either VALUES ?var { OWL ... { Manchester } } or FILTER ( ?var IN ( OWL ... { Manchester } ) ). Prefix rewriting can be used to match endpoint-specific IRIs in systems such as UniProt and BioModels, while external mapping services such as owl:sameAs graphs may be used when available.
Several optimizations are explicitly described:
| Component | Optimization | Function |
|---|---|---|
| Ontology labels | Trie-based indexing | Efficient label search and autocompletion |
| Reasoning | Preclassification with ELK at startup | Accelerates subsumption queries |
| Literature retrieval | Lucene indexes with English analyzer | Tokenization and normalization |
| SPARQL integration | Prefix rewriting | IRI normalization across endpoints |
The constraints are equally central to understanding the framework. ELK supports OWL 2 EL only, and axioms outside EL are ignored, so complex DL constructs not in EL do not contribute to inferences. Ontology heterogeneity remains a practical problem because different URIs or labels for the same relation, such as part_of, part-of, or part of, complicate cross-ontology queries. In the literature module, matching is based on rdfs:label strings only and does not exploit synonyms, acronyms, or specialized named entity recognition, which can reduce recall or precision for some ontologies. Many SPARQL endpoints do not implement OWL entailment regimes, and Aber-OWL: SPARQL compensates by doing reasoning outside the endpoint; however, endpoint evaluation still operates over asserted triples בלבד. No explicit performance benchmarks are reported, even though ELK is designed for large EL ontologies (Hoehndorf et al., 2014).
7. Relation to other resources and significance
Aber-OWL is positioned against several related resources. BioPortal, OLS, and Ontobee expose asserted structure and metadata but do not provide reasoning-based access to inferred ontology content. OntoQuery is a web-based OWL query interface for a single ontology and uses a reasoner, but it lacks multi-ontology support and the literature and data integration provided here. GOAL creates a custom knowledge base blending ontology and annotations and reasons over the combined knowledge base, whereas Aber-OWL separates ontology reasoning from data retrieval, works with multiple ontologies without altering them, and exposes results for query expansion against external endpoints. GoPubMed annotates PubMed with GO and MeSH on the basis of asserted structures, whereas Aber-OWL: PubMed leverages inferences from any ontology to support semantic literature search driven by inferred subclasses or equivalents (Hoehndorf et al., 2014).
The significance of the framework lies in this separation of concerns. It computes inferred ontology content at scale within OWL 2 EL, exposes that content through web services and JSON, and applies inferred sets to semantic retrieval in literature and Linked Data without requiring downstream endpoints to implement OWL entailment. This suggests a general architecture for bridging formal ontology knowledge and practical data access: reasoning is centralized, retrieval remains distributed, and interoperability is achieved through inferred named classes rather than through endpoint-native description logic support.