Enhanced drug-drug interaction extraction from biomedical text using deep learning-based sentence representations.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Drug-drug interactions (DDIs) pose significant risks to patient safety and increase healthcare costs.
• Traditional machine learning (ML) methods struggle with the complexity of DDI extraction from biomedical text.
• Advanced deep learning and transformer models offer better insights but are computationally demanding.

Purpose of the Study:

• To develop an efficient Convolutional Neural Network (CNN) model, named CNN-DDI, for extracting DDIs from biomedical literature.
• To compare the performance of CNN-DDI against traditional ML models and state-of-the-art transformer-based models.

Main Methods:

• Comparative analysis using the SemEval-2013 dataset.
• Evaluation of CNN-DDI against various ML models (Logistic Regression, SVM, Random Forest, Naive Bayes, Decision Trees) and transformer models (BioBERT, RoBERTa, DeBERTa, ELECTRA, DistilBERT).
• Standardized parameter tuning and preprocessing procedures for all models.

Main Results:

• CNN-DDI achieved the highest overall accuracy (86.81%) and F1-score (83.81%).
• CNN-DDI outperformed transformer-based models (best F1-score 81.41%) and traditional ML models (best F1-score 77.09%).
• CNN-DDI demonstrated superior performance with significantly fewer computational requirements.

Conclusions:

• CNN-DDI offers a highly effective and computationally efficient solution for DDI extraction in biomedical text mining.
• The model presents a feasible option for large-scale analysis, balancing performance and resource demands.