The open biomedical annotator.

Clément Jonquet, Nigam H. Shah, Mark A. Musen

TL;DR

This paper introduces the Open Biomedical Annotator (OBA), a web service that automatically tags public biomedical datasets with concepts from established ontologies. By leveraging existing ontologies and semantic expansion techniques, OBA aims to overcome the manual annotation bottleneck, making vast amounts of biomedical data more discoverable and interoperable for translational research.

What Problem Does This Paper Address?

The explosion of publicly available biomedical data presents a significant challenge for researchers: how to effectively find and integrate the specific information they need. While ontologies and terminologies are crucial for structuring and annotating this data, the process is often manual, time-consuming, and requires expert knowledge. This lack of automated, user-friendly annotation tools hinders the ability to perform cross-cutting integrative studies and make translational discoveries. The paper addresses this gap by providing a system that automates the annotation of textual metadata with biomedical ontology concepts.

Methods

The OBA web service operates in two main steps:

  1. Concept Recognition: Raw textual metadata from a dataset is fed into a concept recognition tool (Mgrep) along with a dictionary. This dictionary is built by pooling concept names, synonyms, and labels from various biomedical ontologies (UMLS and NCBO BioPortal). Mgrep uses string matching to identify direct annotations.
  2. Semantic Expansion: The direct annotations are then expanded using semantic relationships within the ontologies. This is achieved through three main components:
  3. Is_a Transitive Closure: This component traverses the parent-child hierarchy of an ontology to add broader concepts to the annotations. For example, if "melanoma" is annotated, its parent concepts like "skin tumor" and "neoplasms" are also added.
  4. Semantic Distance: This component uses a notion of concept similarity to find related concepts. For instance, if "melanoma" is annotated, concepts that are siblings in the hierarchy might also be added.
  5. Ontology Mapping: This component leverages existing mappings between different ontologies (e.g., from UMLS) to generate new annotations. If a concept exists in one ontology, OBA can find its equivalent in another.

Users can parameterize the semantic expansion process, choosing which ontologies to use, the maximum expansion level, and the type of semantic distance. The service returns direct and semantically expanded annotations, scored based on context and frequency.

Datasets

The paper mentions using ontologies from:

  • UMLS (Unified Medical Language System): Encompassing over 94 English ontologies at the time of writing.
  • NCBO BioPortal: Including a subset of over 36 ontologies.

These ontologies collectively provide a dictionary of over 2.6 million concepts and 5.1 million terms. The paper also refers to evaluating OBA on:

  • A set of 1,050,000 PubMed citations (title, abstract, and metadata).
  • Various open biomedical resources, including ArrayExpress experiments, clinical trials, and GEO datasets.

Key Insights and Ideas

  • Bridging the Annotation Gap: The core insight is that automating ontology-based annotation is feasible and crucial for unlocking the potential of massive biomedical datasets. OBA demonstrates a practical solution to the manual annotation bottleneck.
  • Leveraging Ontological Semantics: Simply mapping terms isn't enough. OBA's semantic expansion components (is_a closure, semantic distance, mapping) are key to enriching annotations and revealing hidden relationships within the data, which is a significant advantage over simpler annotation tools.
  • Web Service for Integration: Presenting OBA as a web service (SOAP and RESTful) is critical. This allows it to be easily integrated into existing research workflows and programs, promoting wider adoption and utility.
  • Scalability and Accuracy: The use of Mgrep, a novel radix-tree-based concept recognizer, contributes to high accuracy (>95% in recognizing disease names) and efficient matching, making the system practical for large-scale annotation tasks.

Key Results

  • OBA was evaluated on 1,050,000 PubMed citations, generating over 174 million annotations (18% direct, 82% expanded).
  • An average of 160 annotation concepts were generated per citation, with approximately 99% of the set being annotated with at least one concept.
  • The service is publicly available and has been used internally to build an Open Biomedical Resources (OBR) index, allowing users to search for biomedical data annotated with ontology concepts.
  • Current response times are approximately 20-25 seconds for 500 words, with plans for further optimization.

Limitations and Future Work

Limitations:

  • The concept recognizer (Mgrep) does not perform natural language processing techniques like stemming or spell-checking, though the authors note that biomedical terminologies often contain syntactic variants.
  • Not all NCBO BioPortal ontologies are fully usable via their REST web services API.
  • The service response time can vary depending on the selected semantic expansion components.

Future Work:

  • Enhance Concept Recognition: Incorporate natural language processing techniques and potentially recognize "relations" between concepts.
  • Improve Customizability: Allow for more flexible parameterization of the service, including the ontologies used.
  • Develop New Semantic Expansion Components: Explore additional ways to leverage ontological knowledge for more sophisticated annotation expansion.
  • Optimize Performance: Further technical improvements are planned, including keeping dictionaries loaded in memory and pre-computing hierarchy tables to ensure faster response times.

Summary updated

Abstract

The range of publicly available biomedical data is enormous and is expanding fast. This expansion means that researchers now face a hurdle to extracting the data they need from the large numbers of data that are available. Biomedical researchers have turned to ontologies and terminologies to structure and annotate their data with ontology concepts for better search and retrieval. However, this annotation process cannot be easily automated and often requires expert curators. Plus, there is a lack of easy-to-use systems that facilitate the use of ontologies for annotation. This paper presents the Open Biomedical Annotator (OBA), an ontology-based Web service that annotates public datasets with biomedical ontology concepts based on their textual metadata (www.bioontology.org). The biomedical community can use the annotator service to tag datasets automatically with ontology terms (from UMLS and NCBO BioPortal ontologies). Such annotations facilitate translational discoveries by integrating annotated data.[1].
BibTeX
@inproceedings{jonquet2009open,
  author = {Clément Jonquet and Nigam H. Shah and Mark A. Musen},
  title = {{The open biomedical annotator.}},
  booktitle = {PubMed},
  year = {2009},
}

Loading related content...

Figure