Climate-relevant protein folding — methane-eating enzymes, plastic-degrading proteins

Why protein folding belongs on a climate-quantum machine

Two enzyme families sit on the critical path for atmospheric remediation. Methane monooxygenase (MMO) — both the soluble sMMO and the membrane-bound pMMO variants expressed by methanotrophic bacteria — oxidises CH₄ to methanol at ambient temperature using a di-iron or Cu-cluster active site. PETase, MHETase, and the broader cutinase family hydrolyse the ester backbone of polyethylene terephthalate and adjacent polyesters. Both classes share the same computational pain point: the chemistry happens at a metal centre or a strained transition state where classical force fields are unreliable, and the surrounding protein scaffold is too large to treat with high-level ab initio methods.

Density functional theory gives you the active-site energetics if you crop hard enough; molecular dynamics gives you the scaffold motion if you accept a classical Hamiltonian. The interesting biology — how a distant loop motion gates electron transfer at the di-iron core of sMMO, or why an engineered PETase mutant suddenly works at 70 °C — lives precisely in the QM/MM seam where neither method is honest. That seam is where a fault-tolerant quantum processor eventually earns its keep, and where a 100-qubit NISQ machine can already do useful preparatory work.

The methanotroph problem in concrete terms

pMMO is the harder and more interesting target. The active site is a copper cluster — the exact nuclearity (mono-, di-, or tri-copper) is still contested in the literature — embedded in a transmembrane protein. Methane C–H activation at this site has a barrier in the region of 15–20 kcal/mol depending on which mechanism you believe, and the spin state of the copper centre matters: you cannot describe it with a single Slater determinant.

That is a textbook strongly correlated electron problem. The active space you actually need to treat at full configuration-interaction quality is roughly 20–30 spatial orbitals once you include the copper d-shells, the bridging ligands, and the substrate. Classically, CASSCF(20,20) is at the wall; CASSCF(30,30) is not feasible. On a quantum device, the same active space maps to 40–60 logical qubits under Jordan–Wigner, fewer with parity or Bravyi–Kitaev encodings and qubit tapering off Z₂ symmetries.

On a 100-physical-qubit transmon machine with heavy-hex connectivity, that translates to: useful variational quantum eigensolver (VQE) work on 12–20 qubit reduced active spaces in the near term, with the rest of the protein handled by an embedding scheme — DMET or projection-based WFT-in-DFT — running classically. The quantum machine is not solving the whole protein. It is solving the part of the Hamiltonian where classical methods are demonstrably wrong.

Plastic-degrading enzymes and the engineering loop

PETase is a different shape of problem. The active site is a serine-histidine-aspartate triad — light atoms, no transition metals, no strong correlation. You do not need a quantum computer to describe the catalytic step itself; DFT does fine. What you do need help with is the conformational landscape of engineered variants. The ICCG, FAST-PETase, and HotPETase mutational series have shown that single-residue changes far from the active site shift activity and thermostability by orders of magnitude, and the mechanism is collective backbone reorganisation rather than local electronics.

The quantum angle here is sampling, not energetics. Quantum-enhanced sampling — using a quantum processor to prepare Boltzmann-like distributions over conformational substates, or using quantum-assisted generative models trained on MD trajectories — is an open research area where a 100-qubit machine is the right size of instrument. Algorithms in this family include quantum-augmented Boltzmann machines, quantum approximate optimisation for rotamer assignment, and, further out, quantum signal processing for thermal state preparation. None of these are solved problems. All of them are tractable to investigate at the 50–100 qubit scale, which is exactly the regime Ireland Quantum 100 will sit in.

How the workload actually runs on a transmon device

The machine is a 100-physical-qubit superconducting transmon array, dilution refrigerator at sub-15 mK, heavy-hex coupling topology, no logical error correction in the first cohort window — surface-code logical qubits are on the longer roadmap, not on the Q2 2027 first-customer-access milestone. That has direct consequences for chemistry workloads.

• Circuit depth is the binding constraint. Two-qubit gate fidelities on current-generation transmons sit in the 99.0–99.7% range. A VQE ansatz for a 12-qubit active space with a hardware-efficient or UCCSD-inspired ansatz will run hundreds to low thousands of two-qubit gates before measurement. Error mitigation — zero-noise extrapolation, probabilistic error cancellation, Clifford data regression — is not optional, it is the toolchain.
• Mapping matters. Heavy-hex is not all-to-all. The fermion-to-qubit encoding has to be co-designed with the qubit graph. We will publish layout benchmarks once we have hardware characterisation data; we will not publish vendor-supplied numbers as if they were ours.
• The SDK surface is standard. Workloads will be expressed in OpenQASM 3 with Qiskit, PennyLane, and Cirq front-ends supported. Chemistry-specific tooling — PySCF for the classical front-end, OpenFermion for second-quantised Hamiltonian construction, and embedding via custom DMET drivers — runs on the classical co-processor cluster co-located with the cryostat.

The honest limits at 100 physical qubits

We are not going to fold a 400-residue methane monooxygenase protein on this machine. We are not going to replace AlphaFold. The class of problem this hardware addresses is the small, strongly correlated active-site fragment embedded in a larger classical calculation, plus exploratory sampling work on conformational ensembles. That is a real and currently underserved niche in climate biotech quantum research, and it is the niche that benefits most from sovereign access — these are long, exploratory runs that do not fit neatly into a per-shot cloud pricing model.

Fault-tolerant chemistry — the regime where you can run quantum phase estimation on an industrial-scale enzyme and trust the answer — needs thousands of logical qubits and is a decade-class problem. We are not pretending otherwise. What the 100-qubit machine does is build the pipeline, the embedding code, the error-mitigation playbooks, and the operator expertise that the fault-