AI Scientist Uncovers Potential Treatment for Leading Cause of Vision Loss.

In a compelling demonstration of artificial intelligence’s growing role in research, a team of scientists at FutureHouse has developed an AI platform that may have identified a novel treatment for dry age-related macular degeneration (dAMD)—the most common cause of vision loss in older adults.

The platform, called Robin, is part of a broader effort to create AI systems that can assist with the full range of scientific discovery. As AI has evolved from simple chatbots to complex systems like AlphaFold and ChatGPT, researchers are now focusing on building AI “scientists” that can help formulate hypotheses, review existing literature, design experiments, and analyze data.

Robin represents this new generation of AI tools. Developed as a multi-agent system, it comprises three components: Crow, Falcon, and Finch. Crow performs broad literature searches, Falcon conducts in-depth reviews, and Finch analyzes data and refines hypotheses. Initially separate tools, they were unified to collaborate on an end-to-end research challenge: identifying a potential therapeutic for dAMD.

dAMD affects more than 80% of people diagnosed with age-related macular degeneration and currently has no effective treatment. The condition leads to the gradual breakdown of retinal pigment epithelium (RPE) cells, impairing vision in millions. The FutureHouse team tasked Robin with identifying a drug candidate that could target a disease mechanism relevant to dAMD.

Through iterative analysis, Robin generated and compared ten disease mechanisms using a bracket-style tournament. The most promising approach involved enhancing RPE cell phagocytosis—a natural cellular process essential for eye health. With that as its target, Falcon evaluated drug candidates, and Finch confirmed through experimental data that a rho kinase (ROCK) inhibitor called Y-27632 significantly boosted RPE phagocytosis in lab-grown cells.

To explore how this drug worked, Robin suggested an RNA-sequencing experiment. The results indicated that Y-27632 increased expression of the ABCA1 gene, which plays a key role in lipid transport and cellular health. Building on this, Robin recommended a second round of candidates, ultimately pinpointing ripasudil—a ROCK inhibitor already approved for glaucoma treatment in Japan—as a potential therapy for dAMD. Follow-up experiments verified ripasudil’s effect in enhancing RPE phagocytosis.

Although ROCK inhibitors have been used in eye care, their role in stimulating phagocytosis had not been a central focus in dAMD research. “Robin identified phagocytosis as a key function, which turned out to be a promising direction,” said Ali Ghareeb, ophthalmologist and scientist at FutureHouse.

Ghareeb emphasized the uniqueness of dAMD compared to its better-known counterpart, wet AMD. While both forms share early damage to RPE cells, they diverge in progression and genetics—making the lack of a treatment for dAMD particularly urgent. Robin’s ability to combine insights across medical disciplines was pivotal, he said, likening it to a cross-specialty expert in drug discovery.

The AI’s findings have sparked interest and debate in the scientific community. Konrad Kording, a neuroscientist at the University of Pennsylvania not involved in the research, questioned whether the AI’s proposed mechanism was truly novel, suggesting that knowledgeable researchers may have eventually arrived at a similar conclusion. However, he acknowledged the potential of tools like Robin to increase research efficiency.

The FutureHouse team maintains that, to their knowledge, no prior research had proposed ROCK inhibitors for dAMD in this specific context. While they stopped short of calling the discovery a "move-37" moment—a term used to describe bold, unforeseen moves made by AI—they believe it would have been unlikely to reach the same conclusion without Robin’s assistance.

“Some say it seems obvious in hindsight, but that’s a good sign,” said Andrew White, co-founder and head of science at FutureHouse. “If this ends up helping patients, that’s what really matters.”

Source:https://www.the-scientist.com/an-ai-powered-scientist-proposes-a-treatment-for-blindness-73079

This is non-financial/medical advice and made using AI so could be wrong.