Robin AI repurposes ripasudil for dry AMD

FutureHouse’s Robin system identified ripasudil, a glaucoma drug, as a potential treatment for dry age-related macular degeneration after reviewing 551 papers in about 30 minutes, according to a Nature paper highlighted this week by the Information Technology and Innovation Foundation. The work paired AI-led literature synthesis with human-run lab experiments, and the researchers said those experiments validated ripasudil as a promising candidate. The paper describes Robin as a multi-agent system that can generate hypotheses, propose experiments, analyze results and update its next steps within the same workflow. Nature published the study in May 2026, and ITIF summarized the findings in a June 2 post. ### How did Robin get from papers to a drug candidate? Robin’s authors said the system combined literature-search agents with a data-analysis agent to automate the “key intellectual steps” of scientific discovery. In the dry AMD project, Robin first proposed boosting retinal pigment epithelium, or RPE, phagocytosis as a therapeutic strategy. It then evaluated candidate molecules and moved into lab-guided iteration. (itif.org) FutureHouse said Robin used its Crow agent for a broad literature review, Falcon to assess candidate molecules and Finch to analyze experimental data. The company said human researchers carried out the physical experiments, while Robin generated the hypotheses, experiment choices, data analyses and main manuscript figures. (arxiv.org) ### Why was ripasudil the notable hit? Ripasudil is a clinically used rho kinase, or ROCK, inhibitor that is primarily used for glaucoma, and the paper said it had not previously been proposed for dry AMD. Robin first found that another ROCK inhibitor, Y-27632, increased RPE phagocytosis in cell culture, then used that result to propose a second round of candidates. In that next round, FutureHouse said, ripasudil emerged as the new top hit. (futurehouse.org) The Nature abstract said Robin identified and validated ripasudil as a promising therapeutic candidate for dry AMD. ITIF separately said the system confirmed the drug’s efficacy in lab experiments in a proof-of-concept application focused on the disease. ### What did the lab work show beyond the initial screen? The paper said Robin did not stop at naming a candidate drug. (arxiv.org) After the first experiments, the system proposed an RNA-sequencing follow-up to examine why ROCK inhibition might be increasing phagocytosis in RPE cells. The authors reported that analysis of the RNA-seq experiment pointed to upregulation of ABCA1, which they described as a critical lipid efflux pump and a possible novel target. (nature.com) That means the work produced both a repurposed drug lead and a mechanistic clue that researchers can test further. ### Why are people focusing on the “551 papers in 30 minutes” figure? ITIF said Robin processed 551 papers in about 30 minutes, compared with an estimated 540 hours of manual processing. (arxiv.org) Sandra Barbosu of ITIF wrote that the result showed how AI systems can make “non-obvious connections across disparate areas of the scientific literature” during early-stage drug discovery. Nature’s news coverage said teams of AI agents can generate hypotheses, interpret data and suggest ways to develop medicines, placing Robin in a broader wave of multi-agent research tools. The Robin paper itself framed the system as the first AI workflow to autonomously discover and validate a novel therapeutic candidate in an iterative lab-in-the-loop process. ### What happens next in this project? (itif.org) FutureHouse said on May 20, 2025, that Robin’s code, data and full agent trajectories would be released on May 27. The published paper and preprint now provide the main public record of the dry AMD work, including the ripasudil finding and the ABCA1 follow-up signal, for other researchers to examine and build on. (futurehouse.org) (nature.com)