A Scalable Embedding Based Neural Network Method for Discovering Knowledge From Biomedical Literature.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• The rapid growth of biomedical literature necessitates advanced methods for knowledge discovery.
• Classical information retrieval struggles to identify implicit connections within non-interacting literature.
• Existing literature-based discovery (LBD) approaches often lack scalability and struggle with complex associations.

Purpose of the Study:

• To develop a scalable LBD model for uncovering hidden associations between biomedical entities.
• To integrate biomedical knowledge graphs, graph embedding, and deep learning for enhanced discovery.
• To identify not only associations but also the underlying mechanisms of interaction.

Main Methods:

• Constructing a biomedical knowledge graph by extracting relations from abstracts.
• Applying graph embedding techniques to represent entities and relations in a low-dimensional vector space.
• Training a bidirectional Long Short-Term Memory (BLSTM) network on graph embeddings for association detection.

Main Results:

• The proposed model effectively discovers hidden associations between biomedical entities.
• The method successfully reveals the mechanisms underlying these interactions.
• Experimental results demonstrate the model's capability in both open and closed LBD tasks.

Conclusions:

• Integrating knowledge graphs and deep learning offers an effective strategy for LBD.
• The developed model enhances the ability to capture complex, underlying associations in biomedical literature.
• This approach significantly supports scientific research by revealing novel, implicit knowledge.