Knowledge Guided Attention and Graph Convolutional Networks for Chemical-Disease Relation Extraction

Area of Science:

• Biomedical Natural Language Processing
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Manual extraction of chemical-disease relations (CDR) is time-consuming and labor-intensive.
• Prior knowledge from biomedical databases is crucial for accurate relation extraction.
• Integrating deep learning with prior knowledge offers a promising approach for CDR extraction.

Purpose of the Study:

• To propose a novel model, Knowledge Guided Attention and Graph Convolutional Networks (KGAGN), for automated chemical-disease relation extraction.
• To effectively utilize domain knowledge and syntactic dependency information for improved CDR extraction accuracy.

Main Methods:

• KGAGN employs entity and relation embeddings for feature representation and attention mechanisms to capture contextual information.
• Document-level syntactic dependency graphs are constructed and encoded using Graph Convolutional Networks (GCN) for cross-sentence relation extraction.
• The model integrates weighted context features and long-range dependency features, enriched with prior knowledge, for final relation extraction.

Main Results:

• The KGAGN model achieved an F1-score of 73.3% on the BioCreative-V CDR dataset.
• This performance surpasses existing state-of-the-art methods for chemical-disease relation extraction.
• The study demonstrates the effectiveness of combining deep learning with prior knowledge and syntactic information.

Conclusions:

• The proposed KGAGN model significantly enhances the accuracy of automatic chemical-disease relation extraction.
• Leveraging both domain knowledge and syntactic structures is key to advancing biomedical relation extraction.
• The developed model and its implementation offer a valuable tool for researchers in the field.