Standing
Scaffolded all 41 active sites in its benchmark, against 16 for previous methods. Active enzymes were identified for three catalytic mechanisms after testing fewer than 96 sequences in each case.
Backbone generation · Baker lab · 2025
RFdiffusion2
Designs scaffolds directly from the geometry of catalytic functional groups, without specifying which sequence positions those residues occupy and without inverse rotamer generation.
In plain language
Enzyme design that starts from the geometry of the chemistry. You describe where the catalytic atoms must sit, and the model finds a protein that puts them there, choosing the residue numbering itself.
One way to picture it
Instead of saying "residue 45 must be a histidine", you say "a histidine nitrogen must sit here, at this angle". The model decides where in the chain that lands.
Commonly misread as
Active in a screen of under 96 designs is a strong result for design, and still a long way from an enzyme you would use industrially.
Distinctions
- Unindexed atomic motifs let the model choose where in the chain a catalytic residue lands.
- The specification is a transition-state geometry rather than a residue list.
- It targets enzyme design specifically, where earlier scaffolding methods mostly failed.
Limits
Catalytic activity in a screen of under 96 designs is a strong result for design, and still far from an optimised enzyme. Reported turnover numbers come from individual case studies.
Weights and code
Checked against the registry, not from memory| Repository | Size | Licence | Note |
|---|---|---|---|
| RosettaCommons/RFdiffusion2 | See repository | See repository | — |