A Gate-Based Quantum Computing Framework for Codon Optimization

AI summary

Overview

Research area: Quantum computing for biological sequence design — specifically, gate-based quantum optimization applied to codon optimization in synthetic biology.

Technical level: Advanced. The paper assumes familiarity with Ising/QUBO formulations, Pauli operator mappings, variational quantum algorithms (VQE, SVQE, QAOA), ansatz design, and superconducting hardware execution via IBM Quantum.

Scope (one sentence): The paper formulates codon optimization as a Pauli-encoded Ising Hamiltonian and benchmarks VQE, SVQE, and QAOA against exact diagonalization, classical optimization, and quantum annealing on a 10-qubit, three-amino-acid problem, including executions on the IBM Quantum backend ibm_berlin.

What This Paper Is About

Most amino acids can be encoded by several synonymous codons, so a single protein corresponds to an enormous number of possible DNA sequences — for a protein of 100 amino acids, the authors note more than 10^50 possible synonymous coding sequences. Codon optimization is the task of choosing among these sequences to satisfy biological objectives such as codon usage, GC content, and avoidance of repeated nucleotides, which makes it a hard combinatorial optimization problem.

Prior quantum work on this problem used quantum annealing and QUBO formulations. This paper instead builds a gate-based framework: it converts the same biologically motivated objective into an Ising Hamiltonian written in Pauli operators, then tests whether variational gate-based algorithms can reproduce the exact optimal solutions.

Key Contributions

1. A Pauli-operator Ising formulation of codon optimization. The biological objective and one-hot codon-selection constraint are expressed as a binary quadratic Hamiltonian and mapped to qubits via q_i = (1 − Z_i)/2, producing a cost Hamiltonian used identically by all three algorithms tested.

2. A unified comparison of three gate-based paradigms. VQE, SVQE, and QAOA are implemented on the same Hamiltonian, with systematic evaluation of ansatz choices, classical optimizers, circuit depths, and initialization schemes.

3. Benchmarking against three classical/reference baselines. Results are compared with exact diagonalization, classical optimization, and a D-Wave quantum annealing sampler implemented following the earlier annealing study (Ocean SDK, EmbeddingComposite, chain strength 30, 6000 reads).

4. Hardware validation on IBM Quantum. SVQE and QAOA are executed on ibm_berlin, with the effects of measurement shot count, dynamical decoupling (DD), and gate twirling (GT) on recovery of the degenerate ground-state set reported.

Main Findings

• The benchmark problem has four degenerate ground states. Exact diagonalization of the 1024 possible bitstrings (of which 32 satisfy the one-hot constraints) yields four optimal coding sequences — GAAGGTCCT, GAAGGTCCA, GAAGGACCA, and GAAGGACCT — all at energy −6.2469.

• Classical and annealing baselines recover the full ground-state set. Classical optimization (0.011104 s) and the D-Wave sampler (1.80697 s) both found all four ground states, with D-Wave sampling probabilities of 0.006500, 0.006167, 0.002833, and 0.004000.

• SVQE and QAOA reproduced all four ground states in simulation. SVQE took 6.817 s with probabilities 0.000346, 0.001564, 0.000013, and 0.000128; QAOA took 131.587 s with probabilities 0.000977, 0.000977, 0.002930, and 0.001953.

• VQE was unstable. VQE was run-dependent in simulator tests and did not consistently recover the ground-state solutions, so it was not advanced to hardware and no representative probability or hardware execution time is reported.

• Hardware SVQE recovered ground states but also populated excited states. On ibm_berlin, a single initialization took 1504.599 s of QPU time and returned three of the four ground states plus two states at −6.2222 (GAGGGTCCT at 0.00050 and GAGGGTCCA at 0.00025). Five random initializations cost 8035.612 s and ten random initializations cost 11614.559 s.

• QAOA hardware results depended strongly on execution settings. With 4096 shots, QAOA found three of the four ground states plus one −6.2222 state (12.411 s). Gate twirling at 4096 shots also failed to produce the complete set (14.137 s). All four ground states appeared with 6000 shots (27.910 s), with 4096 shots plus dynamical decoupling (25.809 s), and with 10000 shots (17.726 s).

• Hardware QAOA was faster than the simulator because its parameters were fixed in advance (β = 0.5, γ = 0.8), avoiding repeated variational circuit evaluations.

• Distinct sampling distributions despite identical solutions. The recovered energies and DNA sequences matched across D-Wave, SVQE, and QAOA, but the sampling probabilities differed because each method generates a different distribution.

Methodology in Plain English

Each possible codon choice is represented by a single binary variable, with the rule that exactly one codon must be picked per amino acid (a "one-hot" constraint). The objective is built from four pieces: a codon-usage preference term, a GC-content term that penalizes deviation from a target GC fraction, a term penalizing repeated nucleotides between neighboring codons, and a penalty term enforcing the one-hot rule. These are combined into a quadratic binary model and rewritten as an energy function over quantum spins, so that low energy means a good coding sequence.

The authors then solve this energy function three ways. VQE uses a parameterized circuit whose parameters a classical optimizer tunes to minimize measured energy. SVQE does the same but draws samples from the circuit and filters them for valid one-hot solutions. QAOA alternates cost and mixer layers, with the paper using a single layer (p = 1). Simulations used Qiskit (version 2.2.x): VQE with a TwoLocal ansatz of Ry rotations and CZ entanglers at reps = 4 and the L-BFGS-B optimizer (up to 30,000 iterations); SVQE with the same ansatz family at reps = 2 and L-BFGS-B (up to 300 iterations); QAOA with StatevectorSampler and COBYLA (up to 300 iterations).

For hardware, SVQE ran on ibm_berlin via IBM Runtime with transpilation at optimization level 1, the Runtime Sampler, COBYLA (up to 50 iterations), and 4000 shots per circuit evaluation; additional runs used five and ten random initializations drawn uniformly from [0, 2π). QAOA hardware runs used fixed parameters and varied shot counts, DD, and GT. The test case is deliberately small — glutamic acid, glycine, and proline, giving two, four, and four synonymous codons respectively, for ten qubits total — so that exact diagonalization is available as a reference.

Why This Matters

Impact on research. This is described as a work in progress that establishes a reproducible gate-based workflow for codon optimization, extending the problem beyond the QUBO/annealing paradigm. The authors argue gate-based devices allow constraints to be encoded directly in the Hamiltonian or circuit rather than all objectives being converted into penalty terms, and offer a path compatible with error correction.

Real-world applications:

• Synthetic biology and metabolic/gene engineering, where coding sequences must be designed for a target host organism.
• Recombinant protein production, where codon choice affects expression and protein folding pathways.
• Vaccine development and mRNA-based therapeutics, where sequence stability and translation kinetics matter.
• Gene therapy and biotechnology workflows requiring organism-specific translational tuning.

Industry relevance. The work sits at the intersection of biopharma and quantum computing, and it uses commercially accessible platforms — IBM Quantum's ibm_berlin through IBM Runtime and D-Wave's Ocean SDK — making the workflow directly transferable to industrial quantum-biology pipelines. The reported QPU times (e.g., 1504.599 s for a single SVQE hardware run) also give a concrete sense of current cost overheads.

Future Directions

• Scaling to longer protein sequences. The authors state preliminary investigations on longer amino acid sequences have been initiated and will be reported in future work, together with a comprehensive scalability assessment.
• Testing on next-generation processors. The framework is intended as a scalable methodology to be re-evaluated as quantum hardware improves.
• Better handling of degenerate solution manifolds on hardware. Since shot count, dynamical decoupling, and gate twirling changed whether all four ground states appeared, further work is needed on execution configurations that reliably sample the full degenerate set.
• Resolving VQE's instability. VQE failed to consistently recover ground states at the simulator level; identifying more robust ansatz, optimizer, and initialization combinations remains open.

Target Audience

Quantum algorithm researchers working on combinatorial optimization and variational methods; quantum-biology and bioinformatics groups interested in sequence design; synthetic biology and bioprocessing teams curious about computational design tools; and industry practitioners evaluating near-term gate-based hardware (IBM Quantum, D-Wave) for real optimization workloads. Readers seeking large-scale biological results should note that the demonstrated problem is a 10-qubit, three-amino-acid benchmark and that the authors position the work as an initial framework rather than a production method.

Authors’ abstract

Codon optimization is a challenging combinatorial optimization problem with important applications in synthetic biology, protein expression, and biotechnology. While quantum annealing has previously been explored for this problem, the application of gate-based quantum algorithms remains largely unexplored. In this work, we present a gate-based quantum computing framework for codon optimization by formulating the optimization objective as an Ising Hamiltonian and expressing it in terms of Pauli operators suitable for gate-based quantum processors. The proposed framework is investigated using the Variational Quantum Eigensolver (VQE), Sampling Variational Quantum Eigensolver (SVQE), and the Quantum Approximate Optimization Algorithm (QAOA), with benchmarking against exact diagonalization, classical optimization, and quantum annealing. Using a three-amino-acid benchmark, the proposed implementations successfully reproduce the exact ground-state solutions while identifying optimized ansatz selections, optimization strategies, initialization schemes, and hardware execution settings that provide reproducible performance on current noisy quantum hardware. Hardware demonstrations are performed on IBM Quantum processors to evaluate the practical implementation of the proposed workflow. As a work in progress, this study establishes a reproducible computational framework for gate-based quantum codon optimization and provides the foundation for future investigations of larger protein sequences as quantum hardware and algorithms continue to advance.