AI agents and wet-lab feedback improve aptamer designs over three rounds

In a closed-loop discovery campaign, agent-designed aptamers against thrombin climbed from a 0% hit rate to 42.1% after iterative experimental retraining.

Big Tech
Zachary Cutts · John Coller · Rachel Chen · Hua Tang · Mathias Bellaiche · Ryder Easterlin · +6 more

Google Research · Stanford University · Google Cloud · Google DeepMind

Research Digest··2 min read
The authors ran three cycles of AI-agent-driven aptamer design against α-thrombin and myoglobin, with wet-lab screens feeding results back into the agents.

The authors used single-stranded DNA aptamers as a testbed for closed-loop agentic discovery.

§