Language models for drug-drug interactions: current applications, pitfalls, and future directions

Area of Science:

• Pharmacovigilance and Drug Safety
• Artificial Intelligence in Medicine
• Biomedical Informatics

Background:

• Advanced artificial intelligence (AI) frameworks, particularly large language models (LLMs), are increasingly used for drug-drug interaction (DDI) tasks.
• A significant gap exists in comprehensive reviews detailing LLM applications for identifying known and novel DDIs.

Purpose of the Study:

• To review the current state of LLM-based DDI extraction and prediction.
• To identify methodologies and challenges in applying LLMs to DDI research.

Main Methods:

• A broad literature search was conducted across major scientific databases (PubMed, Embase, Web of Science, Scopus, etc.) from January 2000 to February 2025.
• Methods for DDI extraction using transformer-based models (e.g., BioBERT, GPT) and for DDI prediction using hybrid models, conversational agents, and prompt-based methods were analyzed.

Main Results:

• LLMs demonstrate potential for automating DDI extraction from biomedical text and databases.
• Various LLM architectures and prediction frameworks are being explored for DDI identification.

Conclusions:

• LLMs offer significant potential for advancing pharmacovigilance and clinical decision support systems.
• Addressing limitations in model explainability, reliability (hallucinations), and data quality is crucial for clinical adoption.
• Future research should focus on clinical validation, explainable AI (XAI), data curation, and multimodal data integration.