De novo enzymes with native-like efficiency (kcat/KM ≈ 10⁵ M⁻¹ s⁻¹) without directed evolution
Designed enzymes now reach about 2e5 M^-1 s^-1 on an activated ester (Science 2025) but only about 300 M^-1 s^-1 on retro-aldol, and native-like efficiency across non-activated, multi-step chemistries needs ensemble-level physics not yet available; rated above ensemble prediction (64) because it requires it and more. Application value 52 for industrial biocatalysis (below antibodies at 78), leverage 41 as a proving ground for functional design.
The problem
Designing brand-new enzymes on a computer that work as fast as natural ones, with no lab evolution.
- 1Choose the reaction
- 2Place reactive atoms
- 3AI builds a protein around them
- 4Lab measures speed
- 5Aim: natural speed in one round
Efficiency scale. The usual measure is (units Ms), which describes the speed of the reaction at low substrate concentration. A typical natural enzyme has a of about (and s), with the best near the diffusion limit of to .
Built from
earlier problems → this oneThe problems this one is built from: progress or tools it needs (builds on), special cases that come first (generalizes), and weaker results on the way (implies). On the right, what builds on it.
- Start with
- One classical force field accurate for folded proteins, disordered proteins and weak protein–protein associationComputational Chemistry · D51.5
- Predicting fold-switching proteins and both of their folds from sequenceComputational Medicine · D58
- Step 2
- Boltzmann-weighted conformational ensembles of multi-state proteins from sequenceComputational Medicine · D64
- Then
- De novo enzymes with native-like efficiency (kcat/KM ≈ 10⁵ M⁻¹ s⁻¹) without directed evolutionThis problem
Path to understanding
foundations → problemThe research entries this problem rests on, in the order to read them. Each links to its explainer.
- Start with
- Reverse-mode automatic differentiationAlgorithm
- Fast matrix multiplicationAlgorithm
- Step 2
- The transformer architectureAlgorithm
- Step 3
- Step 4
- Protein language models and generative protein designFrontier · host
- Then
- De novo enzymes with native-like efficiency (kcat/KM ≈ 10⁵ M⁻¹ s⁻¹) without directed evolutionThe problem
State of the art
- Serine hydrolases (Science 2025). Lauko et al. combined RFdiffusion with an ensemble preorganisation scoring method across reaction steps and tested over 300 designs; the best reached catalytic efficiencies up to with crystal structures matching the models (C-alpha RMSD below 1 Å). The reaction is ester hydrolysis, a chemically simple, activated-substrate benchmark; the work is in the range of the target efficiency but on a single reaction class.
Realistic applications
Efficient designer enzymes would enable biocatalytic routes to pharmaceuticals and fine chemicals, plastic degradation, bioremediation and diagnostics, without a natural scaffold to mine and with the possibility of reactions that biology has not evolved. The methods (catalytic array scaffolding, preorganisation scoring) would also inform modelling of natural enzymes.
What counts as solved
Peer-reviewed reports, from at least two independent groups, of single-round computational designs meeting criteria (i)-(iv) above for at least five distinct reactions including two non-activated or multi-step ones, with deposited structures and raw kinetic data.
How it gets verified
ObservationSettled by new measurements or observations.
Needs new wet-lab data: at least two independent groups must report single-round computational designs meeting criteria (i) to (iv) (native-like kcat/KM of about 1e5 per M per s without directed evolution) for at least five distinct reactions including two non-activated or multi-step ones, with deposited structures and raw kinetic data.
Claims under verification
+ add a claimNo claimed results on the Verification board for this problem yet.
Where the work is
Design needs better catalysis models, but the test loop is wet-lab kinetics and structures, so experiment throughput matters as much as modelling.
Bottlenecks
ranked by votes- #1TheoryTransition-state preorganisation to about 1 kcal/mol
Precise placement of catalytic residues did not correlate with activity in Riff-Diff designs; electrostatic preorganisation and dynamics are not captured by scoring.
Community grades
Sign in to add your own grade.
Connected problems
Progress
- Dec 22, 2025Riff-Diff in Naturesource
One-round design of retro-aldolases (91% active, kcat/KM 290) and MBHases approaching evolved enzymes on kcat.
- Apr 18, 2025Designed serine hydrolasessource
RFdiffusion plus preorganisation scoring: over 300 designs tested, kcat/KM up to 2.2e5 M^-1 s^-1, structures within 1 Å.
- Mar 19, 2008Computationally designed Kemp eliminasessource
Rosetta-designed catalysts accelerate Kemp elimination, with low initial efficiency requiring later evolution.