From Code to Catalyst: AI’s Quantum Leap in Designing Life‑Like Enzymes

Imagine a world where plastic litter vanishes not through incineration or landfill burial, but via bespoke enzymes tailored on‑demand to eat away persistent polymers under ambient conditions. Envision pharmaceutical companies spinning up custom biocatalysts in days, sidestepping lengthy directed‑evolution campaigns. This is the near future unveiled by a landmark February 2025 study in Science, where David Baker and colleagues reported the first truly de novo design of serine hydrolases — enzymes that orchestrate four‑step chemical reactions with atomic‑level precision — powered entirely by artificial intelligence.

For decades, protein engineers have wrestled with nature’s complex three‑dimensional folding code, striving to redesign or repurpose existing scaffolds via directed evolution or template‑based computational methods. Yet multi‑step catalysis, which demands precise positioning of catalytic residues at each mechanistic intermediate, remained elusive. The new pipeline — marrying RFdiffusion, a state‑of‑the‑art generative backbone model, with PLACER, an ensemble‑based active‑site evaluator — shatters that barrier. The resulting enzymes, none found in nature, match or exceed the efficiency of their naturally evolved counterparts, marking a watershed moment in synthetic biology.