Extracting microRNA-gene relations from biomedical literature using distant supervision

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genomics

Background:

• Biomedical relation extraction often relies on supervised machine learning, necessitating extensive annotated corpora.
• Distant supervision offers a solution by leveraging knowledge bases to train classifiers, reducing manual annotation efforts.
• MicroRNA-gene relations are crucial for understanding human diseases, but annotated data for their extraction is limited.

Purpose of the Study:

• To develop and evaluate an effective method for extracting microRNA-gene relations from text.
• To address the challenge of limited annotated corpora in biomedical relation extraction.
• To demonstrate the utility of the proposed method in a specific disease context, such as cystic fibrosis.

Main Methods:

• The study proposes IBRel, a novel method based on distantly supervised multi-instance learning for relation extraction.
• IBRel was evaluated on three datasets and compared against co-occurrence and supervised machine learning approaches.
• The method's performance was assessed using F-score metrics.

Main Results:

• IBRel achieved a substantial improvement in F-score (28.3 percentage points higher) on a dataset lacking a specific training set compared to supervised methods.
• While supervised learning outperformed on datasets with specific training sets, IBRel demonstrated strong performance in low-data scenarios.
• The method successfully extracted 27 miRNA-gene relations from recent cystic fibrosis literature.

Conclusions:

• IBRel is a viable and effective approach for extracting microRNA-gene relations from biomedical literature, particularly when annotated corpora are unavailable.
• The method demonstrates the potential of distant supervision and multi-instance learning in advancing biomedical text mining.
• IBRel offers a valuable tool for researchers investigating microRNA-gene interactions in various biological processes and diseases.