Robin automates drug repurposing with lab-in-the-loop experiments and finds 2x cell-level effects

The Batch (Andrew Ng) · rss · 2026-07-22

What happened

FutureHouse, the University of Oxford, and Fordham University released Robin, an open-source agent that proposes repurposed drugs for a given disease by iterating through literature search, mechanism discovery, experiment design, and result analysis.

How it works

It starts from a disease name and generates questions about likely mechanisms.
A literature-search agent summarizes relevant papers and helps Robin identify candidate mechanisms.
Robin designs experiments, ranks them with Claude 3.7 Sonnet, and then searches for existing drugs that may affect the mechanism.
Humans run the lab experiments, upload results, and Robin analyzes the data with a dedicated analysis agent.

Results on dry AMD

In the first run, Robin proposed Y-27632, and human tests on eye cells showed nearly a 2x increase in RPE phagocytosis.
In a second run, it proposed Ripasudil, a glaucoma drug approved in Japan, which increased RPE phagocytosis by 1.89x in the experiment and 1.75x in the follow-up analysis.
The authors emphasize that the drugs were only tested on cells, not in patients.

Why it matters

The pipeline does more than suggest hypotheses: it also analyzes new experimental data and updates its next round of ideas, pointing to a path where agentic systems streamline biomedical discovery.

Original post →

Robin automates drug repurposing with lab-in-the-loop experiments and finds 2x cell-level effects

More from Apps