Out of the dark. Into the reactor.
We search public genomes and metagenomes for enzymes that already work where industrial chemistry struggles — in the cold, at high pH, against C–F bonds — and test every candidate in explicit-solvent simulation before it reaches a lab.
Of the five leads, the two ice-nucleation candidates are the ones leaving the computer: a supplier is being engaged for the bacterial trimer, and the Fusarium ice press peptide has been ordered. Nothing has been received and nothing has been measured. The other three leads are computational.
Lead candidates
Each lead carries two separate facts: a development stage, which says where the molecule is, and a patent status, which says what has been filed. The bacterial INP trimer is at sourcing synthesis and the Fusarium ice press at synthesis ordered; the PFAS defluorinase, the Arctic PETase and the DAC pair are computational. Separately, three U.S. provisional applications cover the ice-nucleation work and nothing is filed for the other three. Every structure here is modeled and simulated in explicit-solvent OpenMM, and each lead is available under a 90-day evaluation option.
What the stage badges mean
A development stage says where the molecule is. It is a separate fact from the patent status, which says what has been filed.
- Computational
- Sequence found and screened; structure predicted; dynamics simulated. No physical material.
- Sourcing synthesis
- A supplier is being engaged; not yet ordered.
- Synthesis ordered
- Material ordered from a supplier; not yet received.
- In hand
- Material received, not yet assayed. Not used yet.
- Measured
- Laboratory data exists. Not used yet.
Soluble bacterial INP trimer
A C3-symmetric β-solenoid homotrimer with its native C-terminal cap. The model presents three ice-binding faces 120° apart and has no transmembrane anchor, so it is designed to stay soluble rather than aggregate.
| Market | $2.5B cold chain & lyophilization |
|---|---|
| Conditions | Predicted onset −1.7 to −2.5 °C |
| MD stability | Equilibrated in explicit TIP4P/Ice water (OpenMM) |
| Active site | Three Thr/Ser/Asn ice-binding faces |
| Expression | pET-28a(+) in E. coli BL21(DE3) [1,857 bp] |
| Stop codons | Tandem TAA-TGA |
Why we start from wild sequence
The enzymes industry needs most evolved under conditions lab strains never faced.
Lab strains hit a ceiling
Most enzyme engineering starts from proteins in E. coli and yeast, which never evolved against 116 kcal/mol C–F bonds, high salinity or sub-zero temperatures. Directed evolution can close some of that gap, but it is slow and expensive.
Designed proteins often fail to express
De novo diffusion models produce plausible backbones without an evolutionary record of folding and expressing in a cell. Many end up in inclusion bodies or unfold under process conditions.
Our approach
Microbes in extreme environments already carry enzymes shaped by that chemistry. We find them in public genomes and environmental metagenomes, and check each active site in explicit-solvent OpenMM simulation before any wet-lab spend.
How it works
Three steps from raw sequence to a lead worth testing.
1. Stream public sequence data
Billions of public sequences in a 2.3 TB archive — mostly environmental metagenomes, plus sequenced isolate genomes — from cold, saline, alkaline and contaminated habitats where the chemistry we need already happens.
2. Filter before folding
A 41-head neural prefilter scores each sequence for every campaign at once on the Apple Neural Engine, then a campaign-specific classifier separates real hits from close look-alikes. 397.8 million distinct sequences have entered the deduplication filter to date, 152.6 million of them screened by Tier 1 in the current pass — on hardware we own. Structure prediction and molecular dynamics run downstream on GPUs, where the throughput limit actually sits.
3. Group by structure, not sequence
Foldseek 3Di clusters proteins by shape, which finds functional homologs below 15% sequence identity to known enzymes. Low identity is a measurement, not a freedom-to-operate opinion.
Also in discovery
From PFAS breakdown to ice nucleation to carbon capture.
Environmental remediation
PFAS defluorinases, PET-degrading enzymes, lanmodulin for rare-earth recovery, heavy-metal binding, marine adhesives, xenobiotic degradation.
Climate & biomineralization
Alkaline-stable carbonic anhydrases for DAC, ice-nucleation proteins, biomineralization scaffolds, carboxysome shells, magnetosomes, pMMO.
Industrial enzymes & materials
Silk fibers, lichen PKS synthases, LPMOs for biomass, polyamide/PU-degrading enzymes, conductive pili, barnacle cement.
Stress protection, sensors & food
Anhydrobiotic IDPs, stress-protective proteins, LPS endotoxin sensors, rare-sugar isomerases, the Fusarium ice press.