Attention guided capsule networks for chemical-protein interaction extraction

Area of Science:

• Biomedical Informatics
• Computational Biology
• Natural Language Processing

Background:

• Automatic extraction of chemical-protein interactions (CPIs) is vital for biomedical research.
• Existing methods face challenges in capturing complex contextual information in biomedical texts.

Purpose of the Study:

• To propose a novel BERT-based attention-guided capsule network (BERT-Att-Capsule) for enhanced CPI extraction.
• To improve the accuracy and efficiency of identifying CPIs from biomedical literature.

Main Methods:

• Utilizing Bidirectional Encoder Representations from Transformers (BERT) for contextual information.
• Employing capsule networks with dynamic routing for sophisticated feature aggregation.
• Integrating multi-head attention to guide feature weighting for improved performance.

Main Results:

• The BERT-Att-Capsule model demonstrated superior performance on the CHEMPROT corpus compared to state-of-the-art methods.
• The model effectively captures long-range dependencies and bidirectional contextual information.
• Attention-guided capsule networks provided fine-grained aggregation, enhancing extraction accuracy.

Conclusions:

• The proposed BERT-Att-Capsule model offers a significant advancement in automatic CPI extraction.
• This approach holds promise for accelerating drug discovery and precision medicine initiatives.
• The model's ability to handle complex features enhances its applicability in biomedical text mining.