The event
In 2025–2026, David Baker's team (2024 Nobel laureate in Chemistry) pushed generative protein design from "making shapes" to "making functions." Their core method, RFdiffusion2, starts from the active-site geometry of an enzyme derived from quantum chemistry, and de novo designs entirely new zinc metallohydrolases, with catalytic activity and crystal structures confirmed in wet-lab experiments.
The methods paper (a bioRxiv preprint) reports that across 41 benchmark active sites, RFdiffusion2 successfully scaffolded 41/41 in silico, whereas the previous best method managed only 16/41 — a difference that turns "can we design, from scratch, a protein that actually catalyzes?" from an occasional lucky event into repeatable engineering.
Why it matters
1. It raises the ceiling of de novo protein design again. AlphaFold solved "predict structure from sequence" (reading molecules); the real dream is the inverse problem — "write a protein for a given function" (making molecules). Previously, RFdiffusion could already build binders, symmetric oligomers, and enzyme active-site scaffolds; ESM3, with 98 billion parameters, generated esmGFP (only 58% identical to any known natural fluorescent protein, roughly 500 million years of evolutionary distance). But "folds correctly" is just the table stakes — "catalyzes" is the new frontier. This work is the first to feed active-site geometric constraints directly into diffusion-based generation, so designs aim at "getting work done" from the very start.
2. It differs from two other bio×AI lines we recently covered. In August we covered Claude autonomously designing protein binders (14 of 15 targets hit, success rates 22.6%–35.1%, independently validated by Adaptyv/Twist wet labs), and Revel's CMLase (AI-screened 50,000 microbial genomes to design catalytic anti-aging peptides). The Claude line is "a general-purpose large model doing binders cross-domain"; CMLase is "screening and modifying nature's existing enzymes"; RFdiffusion2 is generating enzymes from zero — ones that never existed in nature yet carry a defined catalytic mechanism. Together these form three paradigms: general model / modifying nature / creating from nothing.
3. The method is highly transferable. Using "quantum-chemistry-computed active sites" as design anchors means any enzyme with a clearly defined chemical mechanism — not just hydrolases — could potentially be tackled the same way. Combined with the Baker lab's consistent tradition of open weights and public blueprints, this "design → build → test" prove-it loop is turning synthetic biology from a craft into callable, reusable infrastructure.
One-line takeaway
When AI can draw, from scratch, a catalytic protein according to a "desired chemical mechanism" — and wet-lab experiments nod yes — the design space for biomanufacturing truly opens up. Proteins are no longer inheritances of evolution, but custom-order parts.