Drug drug interaction extraction from biomedical literature using syntax convolutional neural network.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Drug-drug interaction (DDI) detection is crucial for public health.
• Current text mining methods for DDI extraction from literature have limitations.
• Advancing automated DDI detection is essential for drug safety.

Purpose of the Study:

• To develop an advanced text mining method for extracting drug-drug interactions (DDIs).
• To improve the performance of DDI detection in biomedical literature.
• To leverage syntactic information for more accurate DDI extraction.

Main Methods:

• A Syntax Convolutional Neural Network (SCNN) model was developed.
• Novel syntax word embeddings incorporating syntactic, positional, and part-of-speech features were proposed.
• An auto-encoder was used to encode bag-of-words features into dense vectors.
• A hybrid approach combined embedding-based convolutional features with traditional features.

Main Results:

• The SCNN method achieved an F-score of 0.686 on the DDIExtraction 2013 corpus.
• This performance surpasses existing state-of-the-art methods for DDI extraction.
• The proposed syntax word embeddings effectively utilize syntactic information.

Conclusions:

• The SCNN model demonstrates superior performance in extracting DDIs from biomedical literature.
• Incorporating syntactic features significantly enhances DDI detection accuracy.
• This method offers a promising advancement for automated drug-drug interaction identification.