Human-supervised Agentic AI for Hypothesis Generation and Experimental Assistance in Drug Repurposing

Agentic AI system RepurAgent outperforms baseline models by generating more novel, credible drug candidates across multiple stages of drug repurposing, including in leukemia, COVID-19, and rare diseases.

• Why it matters: Effective drug repurposing requires comprehensive support throughout the entire lifecycle, from hypothesis creation to experimental validation, which current tools often lack. Addressing this gap can accelerate discovery and improve candidate selection accuracy.
• What they did: A hierarchical multi-agent AI system was developed, integrating a supervisor, planning, and four specialized sub-agents, with human oversight, episodic memory, and retrieval-augmented generation, validated across three real-world scenarios.
• The result: RepurAgent demonstrated superior performance by identifying high-confidence drug candidates, prioritizing compounds with high predictive accuracy, and flagging confounders, thereby enabling more efficient and reliable drug repurposing workflows.