A sequence labeling framework for extracting drug-protein relations from biomedical literature

Area of Science:

• Biomedical Natural Language Processing
• Computational Biology
• Drug Discovery Informatics

Background:

• Automatic extraction of drug-protein interactions is crucial for drug discovery, repurposing, and biomedical knowledge graph construction.
• The BioCreative VII challenge's DrugProt track aimed to advance drug-protein relation extraction.

Purpose of the Study:

• To develop and evaluate novel frameworks for drug-protein relation extraction.
• To compare the efficacy of sequence labeling versus text classification approaches.
• To improve performance through ensemble methods and advanced language models.

Main Methods:

• Comparison of cutting-edge biomedical pre-trained language models within text classification and sequence labeling frameworks.
• Exploration of ensemble methods, including majority voting, to combine model predictions.
• Implementation of a sequence labeling framework for drug-protein relation extraction.

Main Results:

• The sequence labeling framework demonstrated superior efficiency and performance compared to the text classification framework.
• An ensemble of models using majority voting achieved an F1-score of 0.795 on the official test set.
• The best performing approach, an ensemble of sequence labeling models, reached an F1-score of 0.800.

Conclusions:

• Sequence labeling is a more effective framework for drug-protein relation extraction.
• Ensemble methods significantly enhance the performance of relation extraction models.
• The developed approach achieved state-of-the-art results in the BioCreative VII DrugProt challenge.