Google Details Three AI Projects Aimed at Autonomous Wet Labs

Google Research outlined three distinct AI efforts for laboratory work Friday, Oct. 16, spanning robot-command validation and molecule design as components of a longer-term autonomous lab workflow.

The robot system, VERID (VERified Infrastructure for Discovery Laboratories), builds a digital model of a lab workcell, including equipment arrangement, well volumes and reagent inventories. Built on PyLabRobot, it simulates proposed procedures and checks commands against the modeled setup before they reach hardware. The case studies used dye and deionized-water mixtures. In one self-play campaign, agents’ accuracy at identifying blinded dye mixtures rose to more than five times its initial level.

A separate experiment used aptamers, short DNA molecules that bind proteins, in a design-and-screen cycle. The design-to-screen cycle lasted about 10 days; pooled DNA synthesis and reagents were estimated at $3 per candidate. Bead-corrected screening found no hits among 167 new α-thrombin candidates in round one, 54 of 208 (26.0%) in round two, and 131 of 311 (42.1%) in round three. Among strictly de novo candidates without known thrombin motifs, the third-round bead-corrected rate was 37.1% (106 of 286).

Myoglobin results also varied across rounds. Bead-corrected hit rates were 10.2% (20 of 196), 3.7% (nine of 244) and 33.5% (184 of 549). Specificity fell from 75.0% to 19.1%, then recovered to 72.4% after agents adjusted designs in response to cross-reactive sequences.

A third, separate effort designed protein binders for fibroblast activation protein alpha (FAP). Four enriched clones had apparent dissociation constants of 3.0 to 9.1 nanomolar in binding assays. In a computational comparison, the workflow combining human and AI agents had a 9.33% hit rate, compared with 5.06% for the human-only arm and 6.18% for the agent-only arm. The binding findings are early results and do not establish a tested therapy.

Google research scientists Pooja Rao and Annalisa Pawlosky and software engineer Petar Sirkovic wrote, “Moving toward an autonomous wet lab requires solving these open problems with opportunities for human agency and oversight.”

Google plans to connect VERID’s robot-control system with closed-loop protein discovery. The robot tests and molecular-design experiments remain separate efforts.