Quantinuum: Quantum-Enhanced Protein Structure Prediction

The Challenge

Predicting how a protein folds into its functional three-dimensional shape remains one of the hardest problems in computational biology. Classical methods such as molecular mechanics with generalized Born surface area (MM/GBSA) scoring are fast but sacrifice physical accuracy: they approximate quantum mechanical interactions with empirical force fields that struggle to capture subtle electronic effects governing peptide backbone dihedral preferences. For drug discovery, even small errors in conformational ranking translate directly into failed lead compounds and wasted preclinical investment. Quantinuum set out to test whether near-term quantum hardware could close some of that accuracy gap for short peptide fragments, which are increasingly relevant as a drug modality in their own right.

The Quantum Approach

Quantinuum’s research team used the InQuanto quantum chemistry software platform to construct a fragment-based quantum model of a 12-residue peptide. The workflow decomposed the full peptide into overlapping fragments small enough to fit on the H2 trapped-ion processor, then assembled fragment energies into a full conformational energy surface using an incremental many-body expansion. Each fragment Hamiltonian was solved with a variational quantum eigensolver (VQE) augmented by active-space selection to keep qubit counts tractable.

from inquanto.algorithms import VQEClient
from inquanto.ansatze import ChemicallyAwareUCCSD
from inquanto.mappings import JordanWigner

# Build fragment Hamiltonian for a dipeptide unit
hamiltonian = build_fragment_hamiltonian(residues=["ALA", "GLY"], basis="6-31G*")

# Map fermionic Hamiltonian to qubits
mapper = JordanWigner()
qubit_op = mapper.encode(hamiltonian)

# Define UCCSD ansatz with active space (8 electrons, 8 orbitals)
ansatz = ChemicallyAwareUCCSD(
    n_qubits=16,
    n_electrons=8,
    active_orbitals=list(range(8)),
)

# Run VQE on Quantinuum H2
vqe = VQEClient(
    ansatz=ansatz,
    operator=qubit_op,
    backend="quantinuum.hqs-lt-s2",
    optimizer="BFGS",
    max_iterations=300,
)
result = vqe.run()
print(f"Fragment ground-state energy: {result.energy:.6f} Hartree")

The H2 processor’s high two-qubit gate fidelity (above 99.8%) was critical: UCCSD circuits for even small fragments contain deep layers of entangling gates that would be destroyed by noise on lower-fidelity devices. Thirty-two logical qubits were engaged across the combined fragment set, with results stitched together by the InQuanto orchestration layer running on classical hardware.

Results and Implications

Across 48 sampled conformations of the 12-residue peptide, the quantum-fragment method improved conformational ranking accuracy by 18% relative to MM/GBSA, as measured by Spearman rank correlation against coupled-cluster reference calculations. Critically, the quantum approach correctly identified the native-like low-energy basin that classical scoring ranked fourth. This matters because pharmaceutical campaigns routinely prune compound libraries based on exactly such ranked lists.

The study is an early but concrete sign that trapped-ion hardware has crossed a threshold of practical relevance for fragment-level quantum chemistry. As H-series qubit counts grow and InQuanto’s fragment assembly becomes more automated, similar workflows could be applied to full drug-like small molecules interacting with binding-site residues, a problem scale still beyond current hardware but approaching feasibility within the next few years.