OntoGene web services for biomedical text mining

Fabio Rinaldi, Simon Clematide, Hernani Marques, Tilia Ellendorff, Martin Romacker, Raul Rodriguez-EstebanBMC Bioinformatics 2014 · OpenReview

TL;DR

This paper introduces OntoGene web services, a flexible and accessible platform for biomedical text mining. By leveraging the BioC standard and a robust underlying text mining engine, these services aim to democratize access to advanced information extraction capabilities, facilitating knowledge discovery and database curation in the life sciences.

What Problem Does This Paper Address?

The explosion of biomedical literature presents a significant challenge for researchers and curators trying to extract valuable information. Traditional, monolithic text mining applications are often inflexible and difficult to integrate into existing workflows. This creates a bottleneck for tasks like database curation, drug discovery, and general knowledge exploration. The authors address this by proposing a web service-based architecture that offers advanced text mining capabilities in an accessible, modular, and interoperable manner. This approach aims to bridge the gap between the vast amount of unstructured text data and the structured knowledge residing in biomedical databases.

Methods

The core of the OntoGene system is a text mining pipeline that employs both rule-based and machine-learning approaches for entity recognition and relation extraction. Key steps include:

  • Entity Recognition: Identifying mentions of specific biomedical entities (proteins, genes, drugs, diseases, etc.) using lexical resources derived from life science databases. This allows for the resolution of synonyms and cross-linking to existing knowledge bases.
  • Term Normalization and Disambiguation: Matching terms to their canonical representations and resolving ambiguities, such as when a term can refer to different entity types or have multiple identifiers. This is crucial for accurate extraction.
  • Candidate Interaction Generation: Identifying potential relationships between entities. Initially, this is based on co-occurrence within syntactic units.
  • Syntactic Analysis: Employing a dependency parser to generate a detailed syntactic representation of sentences, which is then used to score and filter candidate interactions.
  • Distant Supervision: Utilizing existing databases (like IntAct or CTD) as a source of "weakly labeled" data to train the system, rather than relying solely on manually annotated corpora, which are often limited in scope.
  • Web Services Architecture: The OntoGene system is exposed as a set of web services, allowing remote access to its text mining capabilities. These services utilize the RESTful approach and accept input and return output in the BioC (Biocuration Core) XML format.

Datasets

The paper mentions the use of several life science databases for deriving lexical resources and for training the system through distant supervision. Specific examples include:

  • UniProt (proteins)
  • Entrez Gene (genes)
  • NCBI Taxonomy (species)
  • IntAct (protein interactions)
  • BioGrid (protein and genetic interactions)
  • PharmGKB (drug-gene-disease relations)
  • CTD (chemical-gene-disease relations)
  • RegulonDB (regulatory interactions in E. coli)

The paper also refers to participation in community-organized evaluation challenges like BioCreative, BioNLP, i2b2, CALBC, CLEF-ER, DDI, and BioASQ, which implicitly involve evaluation against benchmark datasets.

Key Insights and Ideas

  • Web Services for Accessibility: Moving from monolithic applications to web services is a critical step for making advanced text mining tools accessible to a broader audience, especially in the pharmaceutical industry where modularity and integration are key.
  • The Power of BioC: The adoption of the BioC standard is highlighted as a significant enabler for interoperability and data exchange in biomedical text mining, simplifying the integration of different tools and modules.
  • Distant Supervision for Scalability: Leveraging existing databases for distant supervision is a pragmatic approach to training text mining models on a scale that manual annotation cannot match, especially for capturing the vastness of biomedical literature.
  • Assisted Curation: The OntoGene system, particularly through its ODIN interface, emphasizes a hybrid approach combining automated text mining with human expertise for curation, aiming to improve efficiency and accuracy.
  • Dynamic Terminology Adaptation: The system's ability to adapt its terminological resources from databases and ontologies allows for customization and specialization to specific use cases.

Key Results

The paper emphasizes the strong performance of the OntoGene system in various community-organized evaluation challenges, often achieving top-ranked results. While specific quantitative results for the web services themselves are not detailed, the underlying OntoGene platform has demonstrated state-of-the-art performance in tasks such as:

  • Protein-protein interaction extraction (BioCreative 2009).
  • Triage tasks in BioCreative 2012.

The web services provide output in the BioC XML format, which can be inspected via a web interface or processed programmatically, offering flexibility for users.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that while the system aims for high recall, it can introduce noisy concepts that require penalization.
  • The current system primarily uses pre-defined terminologies, with limited user ability to upload custom vocabularies.
  • The BioC standard is still relatively new, and its widespread adoption is an ongoing process.
  • Future Work:
    • The authors envision ODIN evolving into a tool for collaborative curation, where human experts can work together, facilitated by the text mining system.
  • They foresee the possibility of users uploading their own terminologies to the web services.
  • The goal is to further automate repetitive curation tasks, freeing up human curators for more complex challenges.
  • The potential for turning OntoGene's results into semantic representations (e.g., RDF) for integration into knowledge graphs is also mentioned.

Summary updated

Abstract

Text mining services are rapidly becoming a crucial component of various knowledge management pipelines, for example in the process of database curation, or for exploration and enrichment of biomedical data within the pharmaceutical industry. Traditional architectures, based on monolithic applications, do not offer sufficient flexibility for a wide range of use case scenarios, and therefore open architectures, as provided by web services, are attracting increased interest.We present an approach towards providing advanced text mining capabilities through web services, using a recently proposed standard for textual data interchange (BioC). The web services leverage a state-of-the-art platform for text mining (OntoGene) which has been tested in several community-organized evaluation challenges, with top ranked results in several of them.
BibTeX
@article{rinaldi2014ontogene,
  author = {Fabio Rinaldi and Simon Clematide and Hernani Marques and Tilia Ellendorff and Martin Romacker and Raul Rodriguez-Esteban},
  title = {{OntoGene web services for biomedical text mining}},
  journal = {BMC Bioinformatics},
  year = {2014},
  url = {https://openreview.net/forum?id=09_XNpUvGMy},
}
Figure