The Lexicon Builder Web service: Building Custom Lexicons from two hundred Biomedical Ontologies.

Gautam K. Parai, Clément Jonquet, Rong Xu, Mark A. Musen, Nigam H. ShahPubMed 2010

TL;DR

This paper introduces the Lexicon Builder Web service, an automated tool for creating custom biomedical lexicons from a vast collection of ontologies. It addresses the laborious manual process of lexicon creation, offering a more efficient and standardized approach for natural language processing (NLP) in the biomedical domain.

What Problem Does This Paper Address?

Biomedical research relies heavily on natural language processing (NLP) for tasks like information extraction, data annotation, and knowledge discovery. A crucial component of NLP systems is the lexicon, a collection of terms and their associated information. Traditionally, these biomedical lexicons are built manually by domain experts, a process that is time-consuming, expensive, and prone to inconsistencies. Furthermore, the ever-expanding landscape of biomedical ontologies (structured vocabularies) presents a challenge in selecting and integrating relevant terms for lexicon creation. This paper tackles the need for an automated, efficient, and flexible solution to generate high-quality, domain-specific biomedical lexicons from these numerous ontologies.

Methods

The Lexicon Builder Web service employs a three-component workflow:

  • Inclusion: This component allows users to select specific ontologies or branches within ontologies. It can also incorporate synonyms and mapped terms from different ontologies, leveraging existing relationships within the BioPortal knowledge base.
  • Exclusion: This is where the "customization" truly shines. Users can filter terms based on several criteria:
  • Medline Frequency: Terms appearing too frequently or too infrequently in Medline abstracts can be excluded, helping to focus on more informative terms.
  • Syntactic Type: Filters terms based on their predominant grammatical role (e.g., noun phrases).
  • UMLS Semantic Type: Utilizes the Unified Medical Language System (UMLS) semantic types to include or exclude terms belonging to specific categories (e.g., "Neoplastic Process").
  • Stopwords: Excludes common or irrelevant terms.
  • Output: This component allows users to define the format of the generated lexicon (e.g., TXT, CSV, XML) and specify which fields to include, such as term IDs and URIs, facilitating integration into downstream workflows.

The novelty lies in the combination of these inclusion and exclusion criteria, allowing for highly tailored lexicon generation, and its deployment as a web service for easy integration.

Datasets

The Lexicon Builder service utilizes a vast collection of publicly available biomedical ontologies, primarily from:

  • UMLS Metathesaurus: A comprehensive collection of biomedical vocabularies.
  • NCBO BioPortal: A repository of numerous biomedical ontologies in various formats.

The service also leverages Medline abstracts for term frequency analysis. The paper does not introduce new benchmark datasets but rather uses existing, large-scale biomedical knowledge resources.

Key Insights and Ideas

  • Automation is Key: The manual creation of biomedical lexicons is a significant bottleneck. Automating this process with a web service dramatically reduces time and effort.
  • Flexibility through Customization: The power of the Lexicon Builder lies in its granular control over lexicon creation. Users can precisely define what goes into their lexicon, making it highly relevant to their specific NLP tasks.
  • Leveraging Existing Knowledge: By tapping into the rich resources of UMLS and BioPortal, the service avoids reinventing the wheel and benefits from the collective effort of ontology development.
  • Standardization and Sharing: The web service promotes collaboration by providing a standardized way to create and share lexicons, fostering consistency across research groups.
  • The Importance of Context: Filtering terms based on Medline frequency and syntactic types highlights the understanding that not all terms are equally useful for NLP tasks; context and relevance matter.

Key Results

The evaluation demonstrated that the Lexicon Builder service achieves high accuracy and runtime performance. Specifically:

  • The service successfully pulled out all desired concepts from various ontologies without including unwanted ones.
  • Performance statistics showed that lexicon generation times were relatively short, even for large ontologies. For instance, generating a lexicon from SNOMED-CT took 15 minutes and resulted in a lexicon size of 4198 MB (uncompressed) with 262,372 unique concepts.
  • The service is currently being evaluated for several real-world use cases, indicating its practical utility.

Limitations and Future Work

  • Manual Evaluation Difficulty: The authors acknowledge that a perfect evaluation would involve comparing against manually created lexicons, which is practically challenging due to the time and expertise required.
  • Future Work: The paper mentions that in future versions, the service will primarily use BioPortal URIs, as UMLS ontologies were not fully integrated into BioPortal at the time of implementation. The service is also being actively evaluated by external research groups for various applications, suggesting ongoing development and refinement based on user feedback.

Summary updated

Abstract

Domain specific biomedical lexicons are extensively used by researchers for natural language processing tasks. Currently these lexicons are created manually by expert curators and there is a pressing need for automated methods to compile such lexicons. The Lexicon Builder Web service addresses this need and reduces the investment of time and effort involved in lexicon maintenance. The service has three components: Inclusion - selects one or several ontologies (or its branches) and includes preferred names and synonym terms; Exclusion - filters terms based on the term's Medline frequency, syntactic type, UMLS semantic type and match with stopwords; Output - aggregates information, handles compression and output formats. Evaluation demonstrates that the service has high accuracy and runtime performance. It is currently being evaluated for several use cases to establish its utility in biomedical information processing tasks. The Lexicon Builder promotes collaboration, sharing and standardization of lexicons amongst researchers by automating the creation, maintainence and cross referencing of custom lexicons.
BibTeX
@inproceedings{parai2010lexicon,
  author = {Gautam K. Parai and Clément Jonquet and Rong Xu and Mark A. Musen and Nigam H. Shah},
  title = {{The Lexicon Builder Web service: Building Custom Lexicons from two hundred Biomedical Ontologies.}},
  booktitle = {PubMed},
  year = {2010},
}
Figure