Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Quantum Sampling Architecture for Protein Structure Reconstruction on Utility-Scale Hardware

T0 review · 4 major / 6 minor · reviewed 2026-07-10 · grok-4.5

Pith's one-line read QSAD predicts short binding-pocket peptide structures on current quantum hardware by sampling a coarse-grained amino-acid Hamiltonian without iterative optimization, beating AI and VQE baselines.

desk verdict Real Heron R2 runs on 101 pocket peptides with clear RMSD wins over AF3/VQE and same-H classical samplers; the open question is how much the fixed lattice H already favors natives. read the letter →

arxiv 2607.06971 v1 pith:Q4R7IFCU submitted 2026-07-08 cs.ET

classification cs.ET
keywords quantumsamplingproteinstructurepredictionbinding-pocketpeptidescoarse-grainedHamiltoniannon-iterativeevolutionNISQhardwareenergylandscapetetrahedrallattice
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Short peptides in protein binding pockets are hard to predict because they lack the long-sequence patterns that data-driven fold predictors rely on, so the problem reduces to physics-based search of a complex energy landscape. This paper presents QSAD, which encodes that landscape as an amino-acid-level Hamiltonian on a tetrahedral lattice and samples it with non-iterative Hamiltonian evolution rather than variational optimization. Run end-to-end on IBM Heron R2 for 101 peptides of 5–18 residues, QSAD reports lower backbone error than AlphaFold3, language-model and MSA fold predictors, and VQE, while staying usable under noise levels that break iterative quantum methods and finishing far faster than VQE. The same measurement ensemble is used to sketch approximate energy landscapes. The claim is that coarse-grained, one-way quantum sampling is already a practical path for this regime on utility-scale superconducting processors.

What carries the argument

QSAD: an amino-acid-level coarse-grained Hamiltonian (tetrahedral-lattice backbone turns plus implicit side-chain contact, burial, local-propensity, and steric coefficients) sampled by randomized-ansatz state preparation followed by multi-β Trotterized Hamiltonian phase evolution and classical decoding/ranking—not iterative variational optimization.

What would settle it

On a held-out set of pocket peptides with crystal structures, if top-ranked or lowest-energy QSAD lattice conformations systematically exceed ~5 Å Cα RMSD from native while classical search (e.g. simulated annealing) on the identical Hamiltonian recovers near-native states, the claim that quantum sampling of this model is what delivers the accuracy would be falsified.

Watch

Extended reading notes

Core claim

On 101 experimentally resolved binding-pocket peptides (5–18 residues), fully executed on IBM Heron R2, non-iterative multi-β sampling of a fixed amino-acid-level tetrahedral-lattice Hamiltonian yields median Cα RMSD near 2.7 Å and improves accuracy by 27–71% over evaluated AI and quantum baselines, while recovering ground-state energy under noise several times typical hardware rates and reducing mean quantum execution time by about 27× relative to VQE.

Load-bearing premise

A single fixed tetrahedral-lattice energy model with literature-derived contact and burial weights, never tuned per sequence, places true native backbones among the low-energy states that multi-β sampling can actually hit.

Editorial extensions

If this is right

  • Short peptides that lack useful MSA or evolutionary signal can be structure-predicted on existing superconducting processors without waiting for fault-tolerant machines.
  • Non-iterative quantum sampling becomes preferable to VQE for noise-sensitive structure tasks because errors do not compound through an optimizer loop.
  • A single sampling run can supply both a predicted backbone and an approximate multi-basin energy landscape for binding and mutational interpretation.
  • Amino-acid-level encoding keeps qubit demand inside current 156-qubit devices for peptides up to about 18 residues, unlike orbital-level formulations.
  • Fixed physics-encoded Hamiltonians can outperform purely learned predictors precisely when sequence signal is weak.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same coarse-grained sampling recipe may transfer to other short interface or disordered peptides outside the binding-pocket benchmark once lattice validity is checked.
  • Because lattice discretization sets a precision floor near 1.6 Å, hybrid pipelines that lift lattice Cα traces into continuous force fields become the natural next bottleneck, not deeper quantum circuits alone.
  • Predictable batch (non-session) quantum job models may become preferred for production bioinformatics workloads where VQE’s long-lived optimizer sessions are hard to schedule.
  • The reported rise in coverage under depolarizing noise suggests controlled noise could be treated as a deliberate exploration knob rather than only a defect to mitigate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The manuscript introduces QSAD, a quantum–classical pipeline that encodes short binding-pocket peptides (5–18 residues) as a coarse-grained six-term Hamiltonian on a tetrahedral lattice with implicit side-chain coefficients, then samples via non-iterative multi-β Hamiltonian phase evolution (randomized EfficientSU2 ansatz + Trotterized e^{-iβH}) on IBM Heron R2. Classical decoding, diversity-preserving ranking, and all-atom reconstruction yield predicted Cα backbones and approximate energy landscapes. On 101 experimentally resolved peptides the method reports median Cα RMSD ~2.7 Å, pairwise wins against AF3, ColabFold variants, ESMFold, OmegaFold, OpenFold, and VQE (e.g. 95/101 vs AF3, 52/55 vs VQE), ~27× lower mean quantum runtime than VQE, and simulated noise resilience up to ε=2% where VQE degrades. Ablations claim that structural quality is fixed at the lattice-sampling stage and that stratified multi-β sampling beats random/greedy/SA pools on the same Hamiltonian.

Significance. If the attribution holds, the work is a concrete demonstration that utility-scale superconducting hardware can deliver competitive short-peptide structure prediction in a regime where MSA/language-model methods are known to be weak, with practical advantages in runtime predictability and noise tolerance over iterative VQE. Strengths that should be credited include: end-to-end execution of 101 cases on Heron R2 with reported telemetry; stage-wise RMSD flatness showing post-processing does not move Cα; matched-pool classical samplers on the same H; and explicit landscape reconstruction with funnel/basin diagnostics. These are rare for quantum biology claims and make the empirical package falsifiable and useful even if some design choices remain heuristic.

major comments (4)
  1. [§4.1, §5.7] §4.1 and §5.7 (weakest load-bearing link): The central claim attributes near-native Cα geometry primarily to non-iterative multi-β sampling of H, not to classical ranking. Section 5.7 shows flat RMSD after lattice decoding and 94–95% wins vs Random/Greedy/SA under matched pool size, which is necessary but not sufficient. The manuscript never reports (i) the energy of the crystal (or best lattice-aligned crystal) conformation under H, (ii) its rank among valid self-avoiding walks, or (iii) an oracle that returns the lowest-H valid lattice structure independent of sampling. Without these, it remains possible that the fixed MJ-style coefficients already place near-natives in the low-energy tail for these 101 pocket peptides, so that any diverse valid pool plus energy ranking recovers good RMSD. Please add energy-rank / oracle-RMSD diagnostics (at least for N≤10 where enumeration or aggressi
  2. [§4.3.2, Fig. 17] §4.3.2 ranking vs Hamiltonian: Candidate ranking reuses Miyazawa–Jernigan contact energy, burial, and related physics terms that are closely related to H_pair / H_node. Combined with energy-based selection in the classical-sampler comparison (Fig. 17), this creates mild circularity between the model that generates samples and the score that picks the reported structure. Clarify which ranking terms are independent of H, report an ablation that ranks by sampling frequency alone (or by a held-out score), and state whether the reported median 2.51 Å end-to-end RMSD still holds under frequency-only selection.
  3. [§5.6, Table 3, Abstract] §5.6 noise resilience: The claim of tolerating noise “3–5× beyond typical hardware error rates” rests on a single 6-residue (18-qubit) depolarizing simulation (Table 3, Fig. 11). Real-hardware VQE failure is shown for one protein (Fig. 2b), but QSAD’s multi-β ensemble is not re-characterized under calibrated Heron noise models or at N>6. Either extend the noise study to a few longer peptides with device-calibrated noise, or narrow the abstract/claim language to the simulated 6-residue setting and the qualitative real-hardware VQE contrast.
  4. [Abstract, §5.3, Fig. 7] Abstract / §5.3 “27–71% improvement”: The percentage range is not derived transparently from the reported medians/means (e.g. 2.7 vs 4.8 Å vs AF3 is ~44% relative reduction; other baselines differ). State the exact formula (relative median RMSD reduction per baseline? min–max over methods?) and which baseline endpoints produce 27% and 71%, so the headline number is reproducible from Table/Fig. 7.
minor comments (6)
  1. [Title page] Title and running headers appear concatenated without spaces (“QUANTUMSAMPLINGARCHITECTURE…”, “APREPRINT- JULY9, 2026”); fix typesetting.
  2. [§4.2.1, Eq. (5)] Eq. (5): UH(β)=e^{-iβ H_D} e^{-iβ H} with H_D=∑X_j is a transverse-field driver; briefly note relation to QAOA-style mixing and whether the order of factors is intentional.
  3. [Table 1] Table 1 QPU times jump sharply with N; a short note on transpiled two-qubit gate counts (not only ansatz depth) would help readers assess scaling.
  4. [§4.4] §4.4 funnel score and basin count are useful but the 35th-percentile threshold and PCA variance (60–75%) should be sensitivity-checked or justified so landscape claims are not threshold-dependent.
  5. [Discussion, Abstract] Discussion correctly notes the ~1.6 Å lattice ceiling and heuristic β={1,2,3,4}; consider moving a one-sentence statement of the ceiling into the abstract or evaluation so readers do not over-interpret sub-2 Å claims.
  6. [References] References include related prior work by overlapping authors (QDockBank, hybrid quantum-AI frameworks); ensure self-citation is balanced with independent lattice-folding and quantum-protein literature already cited (e.g. Robert et al., Boulebnane et al.).

Circularity Check

2 steps flagged · score 2.0 of 10

Central structure claims are externally scored (crystal Cα RMSD vs AI/VQE); only mild self-definitional framing in the noise study (model H ground state) and non-load-bearing self-citation of QDockBank for VQE baselines.

  1. self definitional [§5.6 Noise Resilience, Table 3 and surrounding text]
    "QSAD recovers the ground-state energy (E∗ =−655.0) at every noise level tested up to ε= 2%. This robustness follows from the sampling architecture: corrupted individual measurements do not compound, because the ensemble averages them out instead of feeding them back into an optimization loop."

    E∗ is defined as the minimum of the same coarse-grained Hamiltonian being sampled (the SAASAS test system). Declaring “ground-state recovery” under noise therefore measures whether the sampler finds low-energy bitstrings of H—the objective H itself defines—not an independent experimental free energy or crystal energy. The VQE comparison on the same H is still informative as an algorithmic stress test, but the success metric is self-definitional relative to the model.

  2. self citation load bearing [§5.1 Experimental Setup (dataset and VQE baseline)]
    "55 cases from the QDockBank [3] benchmark (5–14 residues, 7–97 qubits), which also have published VQE results for direct comparison"

    QDockBank is prior work with overlapping authors (Zhang, Yang, Cheng, Fang, Guan, et al.). It supplies both a large fraction of the evaluation set and the only quantum baseline (VQE). This is a minor, non-load-bearing self-citation for the central claim, which rests primarily on crystal RMSD wins against external AI methods; it does not force the 27–71% accuracy result by construction.

full rationale

QSAD’s headline accuracy claim is not circular: predicted backbones are judged by Cα RMSD against experimentally resolved crystal structures, and the method is compared to independent AI systems (AF3, ColabFold, ESMFold, OmegaFold, OpenFold) that do not share the tetrahedral Hamiltonian. Hamiltonian weights are fixed literature constants with no per-sequence fit to the 101-test set, and §5.7 shows classical samplers on the same H under matched pool size still lose on RMSD, so success is not forced by energy ranking alone. Two minor issues remain and justify score 2 rather than 0: (1) the noise-resilience experiment reports recovery of E* = −655, which is the ground state of the model Hamiltonian by construction, not an external physical energy; (2) VQE baselines and part of the dataset come from QDockBank by overlapping authors. Neither step collapses the main derivation chain (sequence → fixed coarse-grained H → multi-β sampling → lattice Cα → crystal RMSD). No fitted-input-as-prediction, uniqueness-from-authors, or ansatz-smuggled-via-self-citation pattern is load-bearing for the 27–71% accuracy claim.

Assumptions & free parameters 6 free parameters · 6 assumptions · 2 invented entities

The claim rests on standard folding thermodynamics plus a stack of modeling choices: tetrahedral lattice geometry, fixed literature contact/burial coefficients with no per-sequence training, heuristic multi-β non-iterative evolution, and classical diversity+physics ranking. Free parameters are mostly schedule and scoring knobs held constant across all 101 cases rather than fit to test RMSD; invented content is the QSAD pipeline and its specific H factorization, not a new physical particle.

free parameters (6)
  • β evolution set = {1,2,3,4}
    Default {1,2,3,4} chosen heuristically and shared across all proteins; Discussion admits non-optimized schedule.
  • Ansatz seeds S and groups G = S=3, G=10, 2000 shots
    S=3 seeds, G=10 batch groups, 2000 shots/circuit set by authors for the campaign budget.
  • EfficientSU2 depth (reps) = reps=1
    reps=1 linear entanglement fixed for all lengths; not variationally optimized.
  • Hamiltonian penalty/weight coefficients (λ_back, λ_0, α_pair/node/loc/ster) = literature-fixed constants (values not tabulated exhaustively)
    Fixed constants from contact potentials and lattice theory applied uniformly; not re-fit per sequence but still model knobs that define which conformations are low energy.
  • Composite ranking term weights = unspecified relative weights
    Six-term physics score (overlap, MJ energy, Rg, burial, contact order, cross-group consensus) used to pick reported structures; relative weights not derived from first principles in-text.
  • Trotter order/repetitions r = 2nd-order; r not fully specified per N
    Second-order Suzuki–Trotter with unspecified r in main text controls evolution accuracy vs depth.
assumptions (6)
  • domain assumption Native structure corresponds to a free-energy minimum of a sequence-dependent energy function (thermodynamic hypothesis).
    Section 2.1; standard Anfinsen-style premise used to justify Hamiltonian ground-state search.
  • domain assumption A tetrahedral lattice with 2(N−1) turn qubits plus contact qubits adequately represents backbone geometry for 5–18-residue pocket peptides.
    Section 4.1.1; coordination-4 matches sp3 geometry but discretizes continuous dihedrals (precision ceiling ~1.6 Å).
  • ad hoc to paper Physicochemical side-chain effects can be absorbed into fixed sequence-dependent coefficients without explicit side-chain qubits.
    Section 4.1.2 ‘Implicit Side-Chain Model’; primary qubit-saving design choice.
  • ad hoc to paper Non-iterative multi-β phase evolution plus randomized ansatz produces a measurement distribution biased toward low-energy conformations without classical feedback.
    Sections 4.2 and 5.7; supported by ablations but not proven optimal.
  • standard math Second-order Trotterization sufficiently approximates e^{-iβH} for the depths used on Heron R2.
    Section 4.2.1 cites Berry et al.; approximation error not quantified per protein.
  • domain assumption Default AI baselines (AF3, ColabFold, ESMFold, etc.) and published QDockBank VQE results are fair comparators for short pocket peptides.
    Section 5.1; defaults may not be peptide-optimized, affecting the 27–71% improvement claim.
invented entities (2)
  • QSAD pipeline (amino-acid Hamiltonian + multi-β one-way sampling + classical decode/rank/reconstruct)
    purpose: End-to-end structure prediction and approximate energy-landscape projection on utility-scale QPUs.
    Named framework integrating encoding, non-iterative circuits, and post-processing; evaluated as a system rather than a new physical law.
  • Six-term coarse-grained protein Hamiltonian H_back+H_dist+H_pair+H_node+H_loc+H_ster on tetrahedral lattice
    purpose: Map sequence conformational space into a qubit eigenproblem with O(N²) qubits.
    Specific term factorization and contact bipartite rules are paper-defined though built from known lattice/contact ideas.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Sampling Architecture for Protein Structure Reconstruction on Utility-Scale Hardware." pith.science (2026). https://pith.science/paper/Q4R7IFCU

@misc{pith2026260706971,
  author       = {Pith},
  title        = {Pith review of: Quantum Sampling Architecture for Protein Structure Reconstruction on Utility-Scale Hardware},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q4R7IFCU}},
  note         = {Machine review of arXiv:2607.06971}
}
read the original abstract

Predicting the structure of short peptides in protein binding pockets remains difficult because this regime requires physics-based conformational search, yet existing methods do not provide a practical way to carry out that search on current hardware. We present QSAD, a quantum-classical framework that reformulates peptide structure prediction as amino-acid-level Hamiltonian sampling and replaces iterative optimization with non-iterative Hamiltonian evolution. Executed entirely on IBM Heron R2 across 101 binding-pocket peptides (5-18 residues), QSAD improves prediction accuracy by 27-71% over all evaluated AI and quantum baselines while maintaining the lowest variance across tested lengths. QSAD also tolerates noise levels 3-5x beyond typical hardware error rates, where iterative methods fail, and reduces mean quantum execution time by 27x relative to VQE. The sampled ensemble further supports approximate reconstruction of protein energy landscapes. These results establish coarse-grained quantum sampling as a practical computational path for structure prediction in regimes where data-driven methods lack sufficient signal.

Figures

Figures reproduced from arXiv: 2607.06971 by the authors.

Figure 1
Figure 1. Pocket–ligand structure of 6OIM, a KRAS G12C system. This example shows the local binding-pocket [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. VQE convergence trajectories. (a) Simulated VQE on a 6-residue peptide under increasing per-gate [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. End-to-end architecture of QSAD. Left: The input protein sequence is encoded into a coarse-grained Hamiltonian. Center-left: Quantum sampling executes on the IBM Heron R2 processor (156 qubits), producing an ensemble of measured bitstrings. Center-right: Classical post-processing decodes bitstrings into lattice structures (Stage 1), scores and ranks candidates with diversity preservation (Stage 2), and reconstructs … view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Circuit-construction workflow of QSAD. The input amino-acid sequence is encoded to form a coarse-grained [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Transpiled QSAD circuit for protein with 4 residues (5 qubits, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 7
Figure 7. Figure 7: Comparison of QSAD against structure-prediction baselines across 101 proteins. (a)–(g) Pairwise RMSD [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Execution time vs. qubit count. QSAD shows linear growth with bounded runtime (max 2.1 h); VQE shows [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Reconstructed energy landscape for protein 1A0Q (16 residues, 120 qubits, funnel score 0.745). (A) 3-D [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: Reconstructed energy landscape for protein 4Y79 (9 residues, 49 qubits, funnel score 0.261). (A) 3-D fitted [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Noise resilience comparison. (a) Best energy vs. error rate: QSAD consistently recovers the ground truth, [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: Structure overlay for protein 2A14 across three pipeline stages. The C [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: Per-protein RMSD across three pipeline stages (101 proteins). The band is flat: post-processing introduces [PITH_FULL_IMAGE:figures/full_fig_p013_13.png]
Figure 14
Figure 14. Figure 14: Sampling heatmaps for three strategy levels versus the true energy landscape. Panel (a) shows the ground [PITH_FULL_IMAGE:figures/full_fig_p014_14.png]
Figure 15
Figure 15. Figure 15: Cumulative unique conformations discovered vs. total shots. L2 maintains a consistently higher discovery [PITH_FULL_IMAGE:figures/full_fig_p014_15.png]
Figure 16
Figure 16. Figure 16: Candidate pool quality under matched pool-size protocol. (a) Pairwise improvement in mean valid RMSD [PITH_FULL_IMAGE:figures/full_fig_p014_16.png]
Figure 17
Figure 17. Figure 17: End-to-end RMSD after energy-based selection and reconstruction. (a) Cumulative distribution: QSAD [PITH_FULL_IMAGE:figures/full_fig_p015_17.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Quantum Circuit Framework for Protein Ensemble-Level Energetics

    cs.ET 2026-08 reject novelty 6.0 of 10

    A residue-level two-state quantum circuit generates solvation microstate ensembles for two Trp-cage proteins, but the claimed Gibbs-sampling and energy-landscape results lack external validation.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    Leveraging machine learning models for peptide–protein interaction prediction.RSC Chemical Biology, 5(5):401–417, 2024

    Song Yin, Xuenan Mi, and Diwakar Shukla. Leveraging machine learning models for peptide–protein interaction prediction.RSC Chemical Biology, 5(5):401–417, 2024

  2. [2]

    Alpha&esmhfolds: a web server for comparing alphafold2 and esmfold models of the human reference proteome.Journal of Molecular Biology, 436(17):168593, 2024

    Matteo Manfredi, Castrense Savojardo, Georgii Iardukhin, Davide Salomoni, Alessandro Costantini, Pier Luigi Martelli, and Rita Casadio. Alpha&esmhfolds: a web server for comparing alphafold2 and esmfold models of the human reference proteome.Journal of Molecular Biology, 436(17):168593, 2024

  3. [3]

    Qdock- bank: A dataset for ligand docking on protein fragments predicted on utility-level quantum computers

    Yuqi Zhang, Yuxin Yang, Cheng-Chang Lu, Weiwen Jiang, Feixiong Cheng, Bo Fang, and Qiang Guan. Qdock- bank: A dataset for ligand docking on protein fragments predicted on utility-level quantum computers. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, pages 746–761, 2025

  4. [4]

    Describing protein folding kinetics by molecular dynamics simulations

    William C Swope, Jed W Pitera, and Frank Suits. Describing protein folding kinetics by molecular dynamics simulations. 1. theory.The Journal of Physical Chemistry B, 108(21):6571–6581, 2004

  5. [5]

    New monte carlo algorithms for protein folding.Current opinion in structural biology, 9(2):177–183, 1999

    Ulrich HE Hansmann and Yuko Okamoto. New monte carlo algorithms for protein folding.Current opinion in structural biology, 9(2):177–183, 1999

  6. [6]

    Application of enhanced sampling monte carlo methods for high-resolution protein-protein docking in rosetta.PLoS One, 10(6):e0125941, 2015

    Zhe Zhang, Christina EM Schindler, Oliver F Lange, and Martin Zacharias. Application of enhanced sampling monte carlo methods for high-resolution protein-protein docking in rosetta.PLoS One, 10(6):e0125941, 2015

  7. [7]

    Alphafold2 and its applications in the fields of biology and medicine.Signal Transduction and Targeted Therapy, 8(1):115, 2023

    Zhenyu Yang, Xiaoxi Zeng, Yi Zhao, and Runsheng Chen. Alphafold2 and its applications in the fields of biology and medicine.Signal Transduction and Targeted Therapy, 8(1):115, 2023

  8. [8]

    Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J Ballard, Joshua Bambrick, et al. Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

Show all 41 references
  1. [9]

    Quantum algorithms for quantum chemistry and quantum materials science.Chemical reviews, 120(22):12685–12717, 2020

    Bela Bauer, Sergey Bravyi, Mario Motta, and Garnet Kin-Lic Chan. Quantum algorithms for quantum chemistry and quantum materials science.Chemical reviews, 120(22):12685–12717, 2020

  2. [10]

    qubit-adapt-vqe: An adaptive algorithm for constructing hardware-efficient ansätze on a quantum processor.PRX Quantum, 2(2):020310, 2021

    Ho Lun Tang, VO Shkolnikov, George S Barron, Harper R Grimsley, Nicholas J Mayhall, Edwin Barnes, and Sophia E Economou. qubit-adapt-vqe: An adaptive algorithm for constructing hardware-efficient ansätze on a quantum processor.PRX Quantum, 2(2):020310, 2021

  3. [11]

    A quantum framework for protein binding-site structure prediction on utility-level quantum processors.Advanced Science, 13(12):e13641, 2026

    Yuqi Zhang, Yuxin Yang, William Martin, Kingsten Lin, Zixu Wang, Cheng-Chang Lu, Weiwen Jiang, Ruth Nussinov, Joseph Loscalzo, Qiang Guan, et al. A quantum framework for protein binding-site structure prediction on utility-level quantum processors.Advanced Science, 13(12):e13641, 2026

  4. [12]

    A hybrid quantum-ai framework for protein structure prediction on nisq devices.arXiv preprint arXiv:2510.06413, 2025

    Yuqi Zhang, Yuxin Yang, Feixiong Chen, Cheng-Chang Lu, Nima Saeidi, Samuel L V olchenboum, Junhan Zhao, Siwei Chen, Weiwen Jiang, and Qiang Guan. A hybrid quantum-ai framework for protein structure prediction on nisq devices.arXiv preprint arXiv:2510.06413, 2025

  5. [13]

    Chemistry beyond the scale of exact diagonalization on a quantum-centric supercomputer.Science Advances, 11(25):eadu9991, 2025

    Javier Robledo-Moreno, Mario Motta, Holger Haas, Ali Javadi-Abhari, Petar Jurcevic, William Kirby, Simon Martiel, Kunal Sharma, Sandeep Sharma, Tomonori Shirakawa, et al. Chemistry beyond the scale of exact diagonalization on a quantum-centric supercomputer.Science Advances, 1...

  6. [14]

    Quantum power method by a superposition of time-evolved states.PRX Quantum, 2(1):010333, 2021

    Kazuhiro Seki and Seiji Yunoki. Quantum power method by a superposition of time-evolved states.PRX Quantum, 2(1):010333, 2021

  7. [15]

    Protein folding thermodynamics and dynamics: where physics, chemistry, and biology meet.Chemical reviews, 106(5):1559–1588, 2006

    Eugene Shakhnovich. Protein folding thermodynamics and dynamics: where physics, chemistry, and biology meet.Chemical reviews, 106(5):1559–1588, 2006

  8. [16]

    Topology, stability, sequence, and length: defining the determinants of two-state protein folding kinetics.Biochemistry, 39(37):11177–11183, 2000

    Kevin W Plaxco, Kim T Simons, Ingo Ruczinski, and David Baker. Topology, stability, sequence, and length: defining the determinants of two-state protein folding kinetics.Biochemistry, 39(37):11177–11183, 2000

  9. [17]

    Molecular dynamics, monte carlo simulations, and langevin dynamics: a computational review.BioMed research international, 2015(1):183918, 2015

    Eric Paquet and Herna L Viktor. Molecular dynamics, monte carlo simulations, and langevin dynamics: a computational review.BioMed research international, 2015(1):183918, 2015

  10. [18]

    Theory of protein folding: the energy landscape perspective.Annual review of physical chemistry, 48(1):545–600, 1997

    José Nelson Onuchic, Zaida Luthey-Schulten, and Peter G Wolynes. Theory of protein folding: the energy landscape perspective.Annual review of physical chemistry, 48(1):545–600, 1997. 16 APREPRINT- JULY9, 2026

  11. [19]

    The structural basis of peptide-protein binding strategies.Structure, 18(2):188–199, 2010

    Nir London, Dana Movshovitz-Attias, and Ora Schueler-Furman. The structural basis of peptide-protein binding strategies.Structure, 18(2):188–199, 2010

  12. [20]

    The energy landscapes and motions of proteins

    Hans Frauenfelder, Stephen G Sligar, and Peter G Wolynes. The energy landscapes and motions of proteins. Science, 254(5038):1598–1603, 1991

  13. [21]

    Peptide conformational sampling using the quantum approximate optimization algorithm.npj Quantum Information, 9(1):70, 2023

    Sami Boulebnane, Xavier Lucas, Agnes Meyder, Stanislaw Adaszewski, and Ashley Montanaro. Peptide conformational sampling using the quantum approximate optimization algorithm.npj Quantum Information, 9(1):70, 2023

  14. [22]

    The relation between the divergence of sequence and structure in proteins

    Cyrus Chothia and Arthur M Lesk. The relation between the divergence of sequence and structure in proteins. The EMBO journal, 5(4):823–826, 1986

  15. [23]

    Protein folding and misfolding.Nature, 426(6968):884–890, 2003

    Christopher M Dobson. Protein folding and misfolding.Nature, 426(6968):884–890, 2003

  16. [24]

    Resource-efficient quantum algorithm for protein folding.npj Quantum Information, 7(1):38, 2021

    Anton Robert, Panagiotis Kl Barkoutsos, Stefan Woerner, and Ivano Tavernelli. Resource-efficient quantum algorithm for protein folding.npj Quantum Information, 7(1):38, 2021

  17. [25]

    A new substitution matrix for protein sequence searches based on contact frequencies in protein structures.Protein Engineering, Design and Selection, 6(3):267–278, 1993

    Sanzo Miyazawa and Robert L Jernigan. A new substitution matrix for protein sequence searches based on contact frequencies in protein structures.Protein Engineering, Design and Selection, 6(3):267–278, 1993

  18. [26]

    Sanzo Miyazawa and Robert L Jernigan. Residue–residue potentials with a favorable contact pair term and an unfavorable high packing density term, for simulation and threading.Journal of molecular biology, 256(3):623– 644, 1996

  19. [27]

    Theory for the folding and stability of globular proteins.Biochemistry, 24(6):1501–1509, 1985

    Ken A Dill. Theory for the folding and stability of globular proteins.Biochemistry, 24(6):1501–1509, 1985

  20. [28]

    Stereochemistry of polypeptide chain configurations.J

    Gopalasamudram Narayana Ramachandran. Stereochemistry of polypeptide chain configurations.J. Mol. Biol., 7:95–99, 1963

  21. [29]

    Efficient quantum algorithms for simulating sparse hamiltonians.Communications in Mathematical Physics, 270(2):359–371, 2007

    Dominic W Berry, Graeme Ahokas, Richard Cleve, and Barry C Sanders. Efficient quantum algorithms for simulating sparse hamiltonians.Communications in Mathematical Physics, 270(2):359–371, 2007

  22. [30]

    Estimation of effective interresidue contact energies from protein crystal structures: quasi-chemical approximation.Macromolecules, 18(3):534–552, 1985

    Sanzo Miyazawa and Robert L Jernigan. Estimation of effective interresidue contact energies from protein crystal structures: quasi-chemical approximation.Macromolecules, 18(3):534–552, 1985

  23. [31]

    All-atom empirical potential for molecular modeling and dynamics studies of proteins.The journal of physical chemistry B, 102(18):3586–3616, 1998

    Alex D MacKerell Jr, Donald Bashford, MLDR Bellott, Roland Leslie Dunbrack Jr, Jeffrey D Evanseck, Martin J Field, Stefan Fischer, Jiali Gao, Houyang Guo, Sookhee Ha, et al. All-atom empirical potential for molecular modeling and dynamics studies of proteins.The journal of phy...

  24. [32]

    Openmm 8: molecular dynamics simulation with machine learning potentials.The Journal of Physical Chemistry B, 128(1):109–116, 2023

    Peter Eastman, Raimondas Galvelis, Raúl P Peláez, Charlles RA Abreu, Stephen E Farr, Emilio Gallicchio, Anton Gorenko, Michael M Henry, Frank Hu, Jing Huang, et al. Openmm 8: molecular dynamics simulation with machine learning potentials.The Journal of Physical Chemistry B, 12...

  25. [33]

    Pyrosetta: a script-based interface for implementing molecular modeling algorithms using rosetta.Bioinformatics, 26(5):689–691, 2010

    Sidhartha Chaudhury, Sergey Lyskov, and Jeffrey J Gray. Pyrosetta: a script-based interface for implementing molecular modeling algorithms using rosetta.Bioinformatics, 26(5):689–691, 2010

  26. [34]

    The maxwell–boltzmann distribution.Molecular Physics, 103(21-23):2821–2828, 2005

    JS Rowlinson*. The maxwell–boltzmann distribution.Molecular Physics, 103(21-23):2821–2828, 2005

  27. [35]

    The pdbbind database: method- ologies and updates.Journal of medicinal chemistry, 48(12):4111–4119, 2005

    Renxiao Wang, Xueliang Fang, Yipin Lu, Chao-Yie Yang, and Shaomeng Wang. The pdbbind database: method- ologies and updates.Journal of medicinal chemistry, 48(12):4111–4119, 2005

  28. [36]

    Colabfold: making protein folding accessible to all.Nature methods, 19(6):679–682, 2022

    Milot Mirdita, Konstantin Schütze, Yoshitaka Moriwaki, Lim Heo, Sergey Ovchinnikov, and Martin Steinegger. Colabfold: making protein folding accessible to all.Nature methods, 19(6):679–682, 2022

  29. [37]

    Language models of protein sequences at the scale of evolution enable accurate structure prediction.bioRxiv, 2022

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Nikita Smetanin, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, et al. Language models of protein sequences at the scale of evolution enable accurate structure prediction.bioRxiv, 2022

  30. [38]

    Lawrence Zitnick, Jerry Ma, and Rob Fergus

    Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott, C. Lawrence Zitnick, Jerry Ma, and Rob Fergus. Biological structure and function emerge from scaling unsuper- vised learning to 250 million protein sequences.PNAS, 2019

  31. [39]

    High-resolution de novo structure prediction from primary sequence

    Ruidong Wu, Fan Ding, Rui Wang, Rui Shen, Xiwen Zhang, Shitong Luo, Chenpeng Su, Zuofan Wu, Qi Xie, Bonnie Berger, Jianzhu Ma, and Jian Peng. High-resolution de novo structure prediction from primary sequence. bioRxiv, 2022

  32. [40]

    Easy and accurate protein structure prediction using colabfold.Nature Protocols, 20(3):620–642, 2025

    Gyuri Kim, Sewon Lee, Eli Levy Karin, Hyunbin Kim, Yoshitaka Moriwaki, Sergey Ovchinnikov, Martin Steinegger, and Milot Mirdita. Easy and accurate protein structure prediction using colabfold.Nature Protocols, 20(3):620–642, 2025. 17 APREPRINT- JULY9, 2026

  33. [41]

    OpenFold: Retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization.bioRxiv, 2022

    Gustaf Ahdritz, Nazim Bouatta, Christina Floristean, Sachin Kadyan, Qinghui Xia, William Gerecke, Timothy J O’Donnell, Daniel Berenberg, Ian Fisk, Niccolò Zanichelli, Bo Zhang, Arkadiusz Nowaczynski, Bei Wang, Marta M Stepniewska-Dziubinska, Shang Zhang, Adegoke Ojewole, Murat...

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.