An agentic AI system for automated pharmacogenomic recommendation generation

Area of Science:

• Biomedical Informatics
• Pharmacogenomics
• Artificial Intelligence

Background:

• Current pharmacogenomic guideline curation is manual, slow, and limited in scope.
• Tailoring drug therapy requires genetic profile analysis, which is challenging with existing methods.

Purpose of the Study:

• To develop an automated, scalable agentic AI system for generating pharmacogenomic recommendations.
• To improve the efficiency and coverage of clinical pharmacogenomic guideline creation.

Main Methods:

• Utilized large language models (LLMs) guided by structured evidence.
• Developed a modular pipeline to process biomedical literature and FDA drug labels.
• Extracted and aggregated clinically relevant entities to generate gene-drug recommendations.

Main Results:

• Achieved high accuracy (91.9%) in extracting clinically relevant entities.
• Generated phenotype-specific dosing recommendations for gene-drug pairs.
• Expert evaluations showed the AI system outperformed leading LLM baselines in clarity and concordance.

Conclusions:

• Agentic AI can automate end-to-end pharmacogenomic evidence synthesis.
• This approach enables broader population coverage and faster guideline updates.
• The system provides consistent and explainable decision support for personalized medicine.