Extraction of semantic biomedical relations from text using conditional random fields

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• The vast biomedical literature requires automated tools for knowledge extraction.
• Named entity recognition is mature, enabling relation extraction.
• Classifying the type of relation is crucial for deeper understanding.

Purpose of the Study:

• To develop an approach for extracting both the existence and type of semantic relations between biomedical entities.
• To address relation extraction without pre-identified entities, treating entity recognition as a subproblem.

Main Methods:

• Utilized Conditional Random Fields (CRFs) for relation extraction.
• Developed a rich set of textual features for the CRF model.
• Applied the approach to disease-treatment and gene-disease relation identification tasks.

Main Results:

• Achieved competitive performance on disease-treatment and gene-disease relation extraction tasks.
• Successfully extracted a gene-disease network with 34,758 associations from the human GeneRIF database.
• The gene-disease network is publicly available as a machine-readable RDF graph.

Conclusions:

• Extended CRFs for semantic relation annotation in the biomedical domain.
• The approach is generalizable to various biological entities and relation types.
• The GeneRIF database is a valuable resource for text mining, with ongoing work to improve entity detection accuracy.