Chemical-induced disease relation extraction with dependency information and prior knowledge

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Chemical-disease relation (CDR) extraction is vital for biomedical research and healthcare.
• Existing knowledge bases (KBs) are valuable but often overlook prior knowledge.
• Dependency trees offer crucial syntactic and semantic information for relation extraction.

Purpose of the Study:

• To propose a novel Convolutional Attention Network (CAN) for enhanced CDR extraction.
• To investigate the combined utility of prior knowledge and dependency path information.
• To improve the accuracy and efficiency of identifying chemical-disease relationships in biomedical text.

Main Methods:

• Extracting the shortest dependency path (SDP) between chemical and disease entities.
• Applying convolution operations on SDPs to derive deep semantic dependency features.
• Utilizing an attention mechanism to weigh semantic dependency vectors against KB representations.
• Combining weighted dependency and knowledge representations for classification using a softmax layer.

Main Results:

• The proposed CAN achieved comparable performance to state-of-the-art systems on the BioCreative V CDR dataset.
• Both dependency path information and prior knowledge were demonstrated to be critical for effective CDR extraction.
• The integration of these two information sources significantly boosted the model's performance.

Conclusions:

• The novel Convolutional Attention Network (CAN) offers a robust approach for CDR extraction.
• Integrating dependency path information and prior knowledge is a promising strategy for improving biomedical relation extraction.
• This method provides a valuable tool for advancing biomedical research and healthcare applications.