Construction of an annotated corpus to support biomedical information extraction

Area of Science:

• Biomedical Informatics
• Computational Biology
• Natural Language Processing

Background:

• Information Extraction (IE) is crucial for knowledge discovery in biomedical text mining.
• Understanding verb and nominalized verb behavior is key for IE.
• Annotated corpora are valuable for training IE components.

Purpose of the Study:

• To define a novel annotation scheme for sentence-bound gene regulation events.
• To create a comprehensive corpus (GREC) for training IE systems.
• To identify participants and assign semantic roles for gene regulation events.

Main Methods:

• Developed a new annotation scheme for gene regulation events, including verbs and nominalized verbs.
• Created the Gene Regulation Event Corpus (GREC) with 240 annotated MEDLINE abstracts.
• Annotated participants with 13 tailored semantic roles and linked to the Gene Regulation Ontology.

Main Results:

• The GREC corpus includes detailed annotations for gene regulation and expression events.
• Annotation achieved inter-annotator agreement rates between 66% and 90%.
• The annotation scheme uniquely captures a wide range of event arguments in the biomedical field.

Conclusions:

• GREC is a unique biomedical resource annotating core relationships and contextual details.
• GREC supports the development of bio-specific tools and resources, including BioLexicon.
• The corpus is suitable for training IE components like semantic role labellers and is freely available.