Biomedical literature mining: graph kernel-based learning for gene-gene interaction extraction

Area of Science:

• Biomedical informatics
• Machine learning
• Genomics

Background:

• Supervised machine learning for biomedical relationship extraction demands extensive manual annotation.
• Existing methods lack focus on gene-gene interactions, crucial for understanding complex diseases.
• Distant supervision offers a practical alternative by leveraging knowledge bases for automatic annotation.

Purpose of the Study:

• To develop and evaluate an automated method for extracting gene-gene interactions.
• To address the gap in biomedical relationship extraction studies focusing on gene-gene interactions.
• To facilitate the understanding of human complex disease heritability.

Main Methods:

• Utilized the KEGG PATHWAY database for gene-gene interaction information.
• Generated a training dataset from PubMed abstracts.
• Employed a graph kernel method for relationship extraction.
• Implemented a distant supervision approach to eliminate manual labeling.

Main Results:

• Achieved a high F1-score of 0.79 for gene-gene interaction extraction.
• Successfully demonstrated the effectiveness of the distant supervision method.
• Validated the applicability of graph kernel methods in this domain.

Conclusions:

• The developed distant supervision method significantly reduces time and cost associated with manual data annotation.
• The graph kernel-based approach is effective for extracting gene-gene interactions.
• This work is expected to contribute to the advancement of precision medicine.