Measured acceleration · Biology · Chemistry

Autonomous enzyme-engineering platform improves two enzymes in four weeks

A generalized autonomous protein-engineering workflow that improved halide methyltransferase and phytase activity through iterative biofoundry experiments

Summary

A Nature Communications paper reported a generalized autonomous enzyme-engineering platform that combines protein language models, supervised machine learning, and biofoundry automation. In two proof-of-concept campaigns, the system improved At HMT and Ym Phytase over four weekly rounds while screening fewer than 500 variants per enzyme.

AI role

Protein language models and supervised machine-learning models designed variant libraries and selected follow-up variants, while the biofoundry executed construction and screening cycles.

Narrative role

This adds measured wet-lab biology acceleration evidence beyond materials self-driving labs, with concrete cycle-time, variant-count, and activity-improvement metrics from a peer-reviewed source.

Caveat

The demonstrations cover two enzymes with defined assays; broader generality across harder protein-engineering objectives still needs more evidence.