What it is
The team introduced RFdiffusion2, a generative protein-design method that places catalytic residues from active-site geometry alone, without needing preset sequence positions. Starting from quantum-chemistry-derived zinc active sites, they built metallohydrolase enzymes and tested a first batch of 96 designs, whose best candidate reached a catalytic efficiency of 16,000 M-1 s-1. A second round of 96 designs produced three more highly active enzymes, topping out at 53,000 M-1 s-1 with a turnover rate of 1.5 per second. The crystal structure of the most active design closely matched its computational model.
Why it matters
Designed metallohydrolases here reached efficiencies orders of magnitude above prior de novo attempts, and did so directly from the computer without rounds of experimental optimization. That combination, high activity from a small tested set (fewer than 100 candidates per round), points toward making enzymes to order for target reactions. It is a concrete step toward treating catalysts as something you specify and generate rather than discover or evolve.
Underlined numbers link to their source. Every metric and quoted figure is listed under Sources and data below.
Filed underenzyme design, protein design, RFdiffusion2, catalysis, generative AI