Biomedical evidence engineering for data-driven discovery

Area of Science:

• Biomedical Informatics
• Computational Biology
• Precision Medicine

Background:

• Precision medicine generates vast health data, necessitating efficient methods for insight verification.
• Manual verification of data-driven insights against biomedical literature is time-consuming and not scalable.
• Intelligent techniques are required to automate the process of evidence discovery.

Purpose of the Study:

• To introduce a framework for biomedical evidence engineering to automate insight verification.
• To develop efficient modules for retrieving and extracting evidence from biomedical literature.
• To address the limitations of manual verification in the era of big health data.

Main Methods:

• A framework combining biomedical literature retrieval and evidence extraction modules.
• Ensemble methods for state-of-the-art biomedical literature retrieval.
• A BERT-based model for extracting evidence in response to specific queries.
• Creation of a large-scale dataset with 1 million biomedical evidence examples, including 10,000 manually annotated instances.

Main Results:

• The retrieval module achieved state-of-the-art performance.
• A BERT-based model was successfully developed for evidence extraction.
• A comprehensive dataset for biomedical evidence engineering was created and made available.
• The proposed framework offers an intelligent solution for evidence discovery.

Conclusions:

• The developed framework significantly enhances the efficiency of verifying data-driven insights from biomedical literature.
• Automated evidence engineering is essential for advancing precision medicine and data-driven discovery.
• The publicly available dataset will facilitate further research in biomedical evidence extraction and validation.