Hierarchical sequence labeling for extracting BEL statements from biomedical literature

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Extracting bio-entity relations from biomedical literature is crucial for knowledge expansion.
• The BioCreative community has focused on evaluating causal relationship extraction algorithms for Biological Expression Language (BEL).

Purpose of the Study:

• To develop a robust framework for hierarchical relation extraction from biomedical texts.
• To improve the accuracy and efficiency of constructing BEL statements from literature.

Main Methods:

• Mapping sentence-level BEL statements to text segments to create hierarchically tagged training instances.
• Inducing a hierarchical sequence labeling model from the generated training instances.
• Applying word alignment to derive an instance-level corpus from the original sentence-level corpus.

Main Results:

• The proposed method achieved a promising F-measure of 31.6% on the BioCreative V Track 4 test corpus for BEL statement extraction.
• The hierarchical sequence labeling approach demonstrated effective utilization of the training corpus.

Conclusions:

• The developed framework offers a significant advancement in hierarchical relation extraction using sequence labeling.
• The approach has potential for further enhancement with advanced machine learning techniques.