BioMCL-DDI Achieves High DDI Extraction Accuracy
On December 5, 2025, Jia et al. publish in PLoS Computational Biology a meta-contrastive few-shot framework called BioMCL-DDI for drug–drug interaction extraction from biomedical literature. The model jointly optimizes prototype-based classification and supervised contrastive representation learning and removes the need for episodic sampling. Evaluated on DDI-2013, DrugBank, and TAC 2018, it reports F1 scores of 87.80%, 86.00%, and ~74.8%, with code publicly released.
Key Points
- 1Introduces BioMCL-DDI, a meta-contrastive few-shot framework for DDI extraction from biomedical literature
- 2Demonstrates improved representation learning by jointly optimizing prototype classification and supervised contrastive loss
- 3Enables robust few-shot performance, achieving F1 up to 87.80% on DDI-2013 and 86.00% on DrugBank
Scoring Rationale
Strong empirical gains and public code justify high impact; novelty is incremental within few-shot relation extraction literature.
Sources
Public references used for this report.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problems
