Extracting chemical-protein relations using attention-based neural networks.

Area of Science:

• Natural Language Processing
• Bioinformatics
• Computational Biology

Background:

• Relation extraction is a key challenge in natural language processing.
• Accurate identification of chemical-protein interactions is vital for drug discovery and understanding biological pathways.

Purpose of the Study:

• To develop and evaluate deep neural network models for text mining chemical-protein interactions.
• To compare the performance of various deep learning architectures, including CNNs, RNNs, and attention-based RNNs.

Main Methods:

• Investigated multiple deep neural network (DNN) models.
• Implemented convolutional neural networks (CNNs), recurrent neural networks (RNNs), and attention-based RNNs (ATT-RNNs).
• Utilized word-level attention mechanisms for enhanced relation extraction.

Main Results:

• Attention-based RNN models outperformed models without attention.
• The ATT-gated recurrent unit (ATT-GRU) achieved the highest micro-average F1 score of 0.527 on the test set.
• Attention mechanisms effectively identified important trigger words without semantic parsing or feature engineering.

Conclusions:

• Deep learning, particularly ATT-RNNs and ATT-GRUs, demonstrates strong performance in chemical-protein relation extraction.
• Attention mechanisms enhance model interpretability and performance by focusing on relevant textual cues.
• The developed approach offers a robust method for automated text mining of chemical-protein interactions.