A quantum algorithm design choice matters more than hardware, at least for a small CRISPR guide RNA selection problem.
Researchers at arXiv published COMET, a head-to-head comparison of two ways to handle constraints in the Quantum Approximate Optimization Algorithm (QAOA) - the dominant near-term quantum heuristic. The test case: picking one guide RNA per gene across three immune-checkpoint targets (PDCD1, LAG3, and HAVCR2), a combinatorial problem that maps cleanly onto a 12-qubit quantum circuit. The conventional approach adds penalty terms to steer the optimizer away from invalid solutions. COMET's alternative uses an XY-mixer, a circuit structure that makes illegal solutions physically unreachable.
The gap is stark. In simulation, the XY-mixer hit above 95% probability of finding the optimal answer by circuit depth p=3. The three penalty variants tested never broke 6% at any depth. On IBM's ibm_kingston processor, the XY-mixer's real-hardware results stayed close to simulation; the worst-tuned penalty variant drifted by +53.9 energy units. The authors are candid that gate-level noise does partially erode the structural guarantee - that honesty is worth noting in a field prone to overselling quantum results.
The 12-qubit problem is, by the authors' own admission, classically trivial - any laptop solves it instantly. The paper's value is methodological: it gives quantum-biology researchers a concrete, hardware-validated reason to prefer constraint-preserving circuit design over penalty tuning. As quantum hardware scales and biologically relevant problem sizes grow, that design choice could determine whether near-term quantum solvers stay competitive.