Google Research: AI agents run closed-loop wet-lab experiments, with a lab digital twin that blocks unsafe robot commands, aptamer hit rates up to 42% after three rounds, and nanomolar protein binders

Why it matters

【Key points】

Google Research published "Grounding AI agents in biological reality: Building blocks for autonomous wet labs" on Oct 16 at 18:11 UTC, covering three projects that bring AI agents to the lab bench.

First, VERID, built on open-source PyLabRobot, is a live 3D digital twin of a liquid-handling workcell that tracks arm geometry, well volumes, reagents and tip stock. Agents dry-run protocols in simulation in seconds, and unsafe or physically impossible commands are blocked. Second, closed-loop DNA aptamer design: over three design-synthesize-sequence rounds, confirmed on-target hit rates reached 33.5% for myoglobin and 42.1% for α-thrombin, and agents spotted "sticky" off-target binders and rewrote their own design rules. Third, de novo binders for FAP, a receptor overexpressed in solid tumors: excising just the binding pocket sped structure prediction 8.6x, human–agent teams nearly doubled a lone scientist's in-silico hit rate, and the lab confirmed single-digit nanomolar binders.

【Verifiable】

Google Research blog: https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/

【Our read】

1. The point is the closed loop: agents revise hypotheses from real wet-lab readouts, a key step from paper to bench.

2. Scale is small: three aptamer rounds on two targets and one protein target; details live in the linked reports.

3. Safety choices stand out: non-infectious systems with no gain-of-function risk, biosecurity screening of every sequence, and human approval and supervision of runs.