arXiv:2609.31077
Quantum Approximate Optimisation Algorithm for Protein Sidechain Packing
Sebastian O. M. Stewart, Nick Chancellor, Jonte R Hance, Ittoop Vergheese Puthoor
Abstract
Sidechain packing is a critical stage in protein folding, with direct implications for structure-based drug discovery. Google DeepMind's tool, AlphaFold2, predicts protein backbone reliably but sidechain positioning less accurately. Recovering the lowest-energy rotamer assignment over a fixed backbone is NP-hard. We present a hybrid quantum-classical pipeline that repacks sidechains on the AlphaFold backbone using the Quantum Approximate Optimisation Algorithm (QAOA), encoding the one- and two-body energies as a quadratic unconstrained binary optimisation (QUBO) problem. We introduce a constraint-preserving ansatz, pairing a W-state initialisation with a cyclic XY ring mixer, that enforces one-hot rotamer validity without penalty terms while keeping two-qubit gate scaling linear in the rotamer count. We also define an asymptotic shot-scaling metric, measured against the experimentally resolved conformation, that fixes optimiser quality independently of the baseline; its fitted growth stays below the classical exhaustive-search rate at moderate rotamer flexibility. Evaluated on bovine pancreatic trypsin inhibitor (5PTI) across high- and moderate AlphaFold-confidence regions, the pipeline lowers conformational energy against the AlphaFold baseline.
AI Summary
The paper studies fixed-backbone protein sidechain packing with QAOA: selecting one rotamer per residue to minimise a pairwise conformational-energy model. A hybrid pipeline uses PyRosetta to generate and prune rotamers, maps their energies to a one-hot QUBO, and optimises a constraint-preserving circuit. Each residue register starts in a W state; an XY ring mixer preserves one-hot validity without penalties. The mixer uses $O(MN)$ two-qubit gates, versus $O(MN^2)$ for complete-graph mixing, for $M$ residues and $N$ rotamers per residue. State-vector simulations examine subsections of bovine pancreatic trypsin inhibitor (5PTI), using 30 seeds and depths up to $p=12$, with registers up to 22 qubits. Exhaustive enumeration on experimentally resolved-backbone instances defines success as energy within $1.5$ kcal/mol of the discrete minimum. Success probabilities determine the shots needed for $99.99\%$ recovery. An exponential fit grows as $e^{0.31Q}$ in qubit count $Q$ and remains below 500 shots through 22 qubits, although instance-dependent budgets vary substantially. Repacking AlphaFold-backbone regions at 5–14 qubits beats the baseline by more than $1.5$ kcal/mol in $15.6\%$ of high-confidence runs and $54\%$ of moderate-confidence runs; mean improvements among wins are $1.864$ and $8.114$ kcal/mol, respectively. High-confidence wins disappear at 22 qubits. The study is limited to one protein and noiseless simulation. Generalisation of the fitted sampling budget remains open, hardware deployment requires explicit W-state preparation and noise evaluation, and energy improvements are not validated against binding affinity or structural stability. Larger gains against moderate-confidence predictions are attributed to weaker baselines, not stronger quantum optimisation.
Central claim may not hold: Success probabilities calibrated on experimental-backbone Hamiltonians do not guarantee optimal selection on AlphaFold-backbone Hamiltonians or exclude solver failure there. Small-instance shot-budget fits do not establish asymptotic scaling, their success event includes near-optimal assignments rather than only the GMEC, and one-hot encoding does not reduce the feasible state-space dimension relative to classical assignments. The claimed conservative shot-budget bound is already exceeded by a reported 22-qubit subsection requiring approximately 732.88 shots on average, versus a fitted budget below 500. The claimed scaling advantage for heterogeneous rotamer counts does not follow from their average, as equal numbers of one- and five-rotamer residues give an exhaustive-search exponent approximately 0.268 below the fitted shot exponent 0.31. The claimed two-round ring-mixer depth excludes odd rings, and partitioned XY edge gates do not generally implement the exact exponential of the stated summed mixer Hamiltonian.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 6 pages, 3 figures