Google DeepMind’s AlphaProtein Novo designs new-to-nature enzymes from scratch (piperidine synthesis, DEHP degradation), with Frances Arnold’s lab
Event confirmedAwaiting review
Importance: major (4 of 5)The takeaway
On Oct 5, 2026 Google DeepMind posted two bioRxiv preprints on AlphaProtein Novo, a diffusion system that co-generates enzyme structure and sequence around a target reaction.
Status
- Claim
Event confirmedAwaiting review
- Our reporting
- High confidence
- Verification
- Preprint only
- Importance
- Major (4 of 5)
- Last verified
- 7 October 2026
Your AI and this story
- GPT-6.2 Astra127 days after its cutoff
- Claude Opus 5.597 days after its cutoff
- Gemini 4 Hydrogen188 days after its cutoff
- Grok 4.7127 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 97 days before it.
Key facts
- Preprint 1: ‘Designing enzymes for new-to-nature chemistry and non-natural substrates with AlphaProtein Novo’ (Wu, Abramson, Frerix et al.; Google DeepMind, Caltech, Pitt), bioRxiv 10.64898/2026.10.01.756017
- Claim: ‘the first time de novo enzyme design can outperform natural sequence mining’ on challenging reactions; competitive activity on benchmark reactions; key ingredient: mechanism-inspired metrics based on AlphaFold 3 predictions
- Preprint 2 (with Frances Arnold’s Caltech lab and Peng Liu’s Pitt group): ‘Design-Evolution Synergy Enables New-to-Nature Enzyme Chemistry’, de novo hemoproteins for regio- and enantioselective C–H amination to chiral piperidines, combining motif scaffolding, redesign and directed evolution (bioRxiv 10.64898/2026.10.01.754751)
- DEHP (a phthalate plasticizer pollutant) degraders that remain active under denaturing conditions that inactivate natural enzymes
- Repository: google-deepmind/alphaprotein-novo, a diffusion pipeline with LigandMPNN redesign and AlphaFold 3 scoring; benchmarks include Kemp eliminase, serine esterase, DEHPase, carbene and nitrene transferases; code Apache 2.0, weights under separate terms of use
- Announced by Pushmeet Kohli on X (~31k views, Oct 5), crediting Jue Wang, Zachary Wu, Josh Abramson and the Arnold and Liu groups
- Secondary claim (Startup Fortune): one enzyme makes piperidine ‘99 times faster than current industrial methods’; not found in the abstracts, unverified
What happened
DeepMind extended its protein work (AlphaFold for structure, AlphaProteo for binders) to enzymes. AlphaProtein Novo generates a protein’s backbone and sequence together around a catalytic motif, redesigns sequences with LigandMPNN and filters designs with AlphaFold 3-based, mechanism-aware metrics. With Frances Arnold’s lab, designs were further improved by directed evolution into efficient, stable enzymes for chiral piperidine synthesis.
Why it matters
Most industrial enzymes start from natural proteins and are improved by mutation. A generator that beats natural starting points for non-natural chemistry would open enzyme routes to drug intermediates and pollutant clean-up that nature never evolved. Results are preprints; claims such as “99x faster than industrial methods” come from secondary coverage and are not verified here.
What is disputed or not yet verified
| Verification | Preprints (bioRxiv), lab-validated by the authors; not peer-reviewed |
|---|
Sources
5 sources from 4 sites. Numbers match the chips in the text.
5 sources: 4 primary, 1 press
Primary
- bioRxiv: Designing enzymes for new-to-nature chemistry and non-natural substrates with AlphaProtein Novobiorxiv.org, paper
- bioRxiv: Design-Evolution Synergy Enables New-to-Nature Enzyme Chemistrybiorxiv.org, paper
- GitHub: google-deepmind/alphaprotein-novogithub.com, code
- Pushmeet Kohli on X: AlphaProtein Novo preprintx.com, official
Press
- Startup Fortune: DeepMind unveils AI system that designs enzymes no living cell ever madestartupfortune.com, press
Changes
- Science block: kind: other (not a long-open problem; listed with other AI science results, not in the open-problems count)
- Filed