Drug-Drug Interaction Extraction via Recurrent Hybrid Convolutional Neural Networks with an Improved Focal Loss.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Drug-drug interactions (DDIs) pose significant health risks.
• Automatic extraction of DDIs from literature is crucial for drug development and patient safety.

Purpose of the Study:

• To propose a novel Recurrent Hybrid Convolutional Neural Network (RHCNN) for accurate DDI extraction.
• To enhance DDI extraction by addressing class imbalance issues in datasets.

Main Methods:

• Utilized a hybrid approach combining recurrent and convolutional neural networks for feature extraction.
• Incorporated semantic and position embeddings, fusing word embeddings with contextual information.
• Employed an improved focal loss function to handle imbalanced data.

Main Results:

• Achieved a micro F-score of 75.48% on the DDIExtraction 2013 dataset.
• Outperformed the state-of-the-art approach by 2.49% in DDI extraction.

Conclusions:

• The proposed RHCNN model demonstrates superior performance in automatic DDI extraction.
• The method effectively addresses challenges associated with class imbalance in DDI datasets.