Associating biological context with protein-protein interactions through text mining at PubMed scale

Area of Science:

• Biomedical Informatics
• Computational Biology
• Pharmacology

Background:

• Drug repurposing holds significant clinical potential, particularly for rare diseases.
• Accurate biomedical knowledge graphs require biological context (e.g., cell type, tissue).
• Existing knowledge graphs lack essential cell type and tissue information.

Purpose of the Study:

• To develop a method for associating biological context with protein-protein interactions extracted from text.
• To create large-scale, automatically generated corpora for training context association classifiers.
• To enhance text-derived knowledge bases with specific biological details.

Main Methods:

• Framing context association as a classification task using syntactic, semantic, and meta-discourse features.
• Introducing the "Insider corpora" generated from PubMed for classifier training.
• Utilizing precise syntactic cues to identify cell type and tissue relevance in extracted relations.

Main Results:

• Achieved F1 scores of 0.955 for cell type identification and 0.862 for tissue identification.
• Demonstrated that intuitive, interpretable features can effectively address context association.
• Successfully incorporated cell type context into a protein-protein network for dengue fever.

Conclusions:

• The developed framework enables the enrichment of biomedical knowledge graphs with essential biological context.
• This approach facilitates more principled pharmacological discovery and drug repurposing.
• The "Insider corpora" provide a scalable resource for training context-aware biomedical relation extractors.