Design the Chemistry Before the Scaffold

An industrial enzyme team needs a reaction to work under useful conditions. A convincing fold is only part of that problem. The protein must position the substrate, support the chemistry, and keep doing so beyond a single reaction event. The assay endpoint belongs in the brief from the start.

In the RFdiffusion3 study, the enzyme brief was not a vague request for hydrolase activity. The authors specified functional atoms from a Cys-His-Asp catalytic triad, a glutamine, backbone atoms flanking the catalytic cysteine, and a substrate positioned in a tetrahedral-intermediate geometry.

The study screened an activated ester substrate. That is a useful demonstration of designed catalysis, but it does not establish activity on an industrial substrate or performance under process conditions. Those application requirements must shape a separate design brief.

The published workflow is a relay. RFdiffusion3 proposes all-atom scaffolds around the catalytic motif. LigandMPNN proposes sequences for those structures. Structure prediction checks whether a sequence returns to the intended geometry. Each stage filters a different failure mode.

The official RFdiffusion3 implementation supports atom-level constraints, motif scaffolding, ligand context, and hydrogen-bond conditioning. Its enzyme-design tutorial also releases the generated structure used in the walkthrough above. That structure shows the calculation, not catalytic activity, and it is not presented as one of the paper's active enzymes.

The practical decision is whether the proposed geometry is credible enough to justify an assay panel. Agreement between sequence and structure models can support that decision; it cannot establish turnover. The tutorial above illustrates this computational stage only.

The authors prospectively screened 190 designs against an activated ester substrate. Thirty-five showed multiple turnovers. The most active reported design reached a catalytic efficiency of 3,557 M⁻¹ s⁻¹.

Most screened designs did not meet the multi-turnover bar. A program therefore needs both a credible structural shortlist and enough assay capacity to learn from the failures.

The reported hit rate belongs to this reaction, motif, pipeline, and screen. It does not establish a general success rate for enzyme design, another substrate, or a development program with different expression, selectivity, stability, or manufacturing requirements.

The next brief should start with the intended application: the substrate that matters, the conditions the enzyme must tolerate, and the activity that would justify further development. That gives a measured hit a clear next question instead of treating the screen as the finish line.

For a Crane study, we would agree those criteria with the experimental team before choosing a design panel. The useful outcome is a set of candidates with a stated reason to test each one and a plan for interpreting the measurements, including designs that miss the bar.