AI 'Co-Scientist' Speeds Discovery of New Candidates for Macular Degeneration

A new artificial intelligence system named "Robin" has successfully automated the continuous loop of generating hypotheses and analyzing experimental data. Developed by researchers at FutureHouse, the University of Oxford, and Fordham University, Robin demonstrated its capabilities by identifying promising new therapeutic candidates for dry age-related macular degeneration (dAMD), a leading cause of blindness.

To test the system, the research team tasked Robin with finding treatments for dAMD. Robin autonomously scoured the scientific literature, proposed enhancing cellular phagocytosis as a treatment strategy, and suggested several existing drug candidates. When human researchers tested drugs from Robin's first suggestions in the lab, Robin's data analysis agents processed the raw flow cytometry results and confirmed increased phagocytosis. Robin then proposed and analyzed a follow-up RNA-sequencing experiment, which pointed to higher activity of ABCA1, a lipid efflux pump involved in clearing debris in the eye. In later rounds of testing, Robin's analysis identified two distinct compounds—ripasudil, a glaucoma drug, and KL001—that also increased phagocytosis in specialized cells lining the back of the human eye.

This milestone represents the first multi-agent system capable of fully automating both hypothesis generation and data analysis for experimental biology. Traditionally, drug repurposing requires immense human effort to synthesize decades of compartmentalized research. Robin reduces the cognitive labor of a discovery cycle from an estimated 872 to 937 human hours to less than two hours, synthesizing hundreds of scientific papers in minutes.

While Robin's efficiency is remarkable, these findings remain preclinical. The therapeutic candidates for dAMD have only been validated in in vitro cell models and will require rigorous in vivo testing and randomized clinical trials to confirm safety and efficacy. Furthermore, Robin operates as a "lab-in-the-loop" framework, relying on human scientists to physically execute the proposed experiments.

By seamlessly connecting literature-based hypotheses with raw laboratory data analysis, Robin provides a scalable blueprint for accelerating therapeutic development.

Reference:
Ghareeb, A. E. et al. A multi-agent system for automating scientific discovery. Nature (2026). https://doi.org/10.1038/s41586-026-10652-y