MetaTron

MetaTron extracts and structures biomedical annotations to produce machine-readable corpora for relation extraction, topic recognition, and entity linking.

Key Features:

• Mention-level and document-level annotation: Performs both mention-level and document-level annotations to capture entity mentions and document-level labels.
• Integration of automatic predictions: Incorporates automatic built-in predictions for entity and relation identification.
• Ontology-supported relation annotation: Supports relation annotation using ontologies to represent structured relationships among biomedical entities.
• Format and data source support: Supports customization and processing of documents in formats including PDFs and abstracts from PubMed, Semantic Scholar, and OpenAIRE.

Scientific Applications:

• Relation extraction: Generates annotated corpora for supervised relation extraction involving biomedical entities.
• Topic recognition: Produces annotations suitable for topic recognition in biomedical literature.
• Entity linking: Produces entity-level annotations to facilitate linking mentions to ontologies.
• Machine learning dataset creation: Facilitates creation of annotated datasets for training and evaluating machine learning models.

Methodology:

A qualitative analysis compared MetaTron to manual annotation tools TeamTat, INCEpTION, LightTag, MedTAG, and brat based on technical, data, and functional criteria, and a quantitative evaluation assessed MetaTron's performance in terms of time efficiency and number of clicks required for document annotation.