Approximate subgraph matching-based literature mining for biomedical events and relations

Area of Science:

• Biomedical informatics
• Computational biology
• Natural Language Processing

Background:

• Biomedical text mining aims to extract relations and events involving genes/proteins from literature.
• Existing methods face challenges in handling complex dependency contexts.

Purpose of the Study:

• To propose a novel approach for mining relations and events in biomedical literature.
• To improve the extraction of biological knowledge from complex textual data.

Main Methods:

• Utilizing approximate subgraph matching for knowledge extraction.
• Searching for approximate subgraph isomorphism between sentence graphs and contextual dependencies.
• Introducing error tolerance into graph matching to handle complex contexts.

Main Results:

• Achieved a 51.12% F-score for extracting nine types of biological events (BioNLP-ST 2011 GE task).
• Achieved an 84.22% F-score for detecting protein-residue associations.
• Demonstrated high precision and recall in extracting relations and events from complex contexts.

Conclusions:

• The proposed approximate subgraph matching approach is effective for biomedical relation and event extraction.
• The method shows generalizability across different biomedical text mining tasks.
• Error-tolerant graph matching enhances retrieval of knowledge from complex literature.