Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Factorization Machine with Quadratic-Optimization Annealing for RNA Inverse Folding and Evaluation of Binary-Integer Encoding and Nucleotide Assignment

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper shows that in FMQA-based RNA inverse folding, one-hot and domain-wall binary encodings beat binary and unary encodings, and that domain-wall encoding wins outright when guanine and cytosine are assigned to the boundary integers.

desk verdict The encoding/assignment findings are likely real, but the 'fewer evaluations' claim is undermined by tuning the configuration on the same benchmark. read the letter →

arxiv 2602.16643 v1 pith:NRVBTI6A submitted 2026-02-18 cs.LG cond-mat.stat-mech

Shuta Kikuchi , Shu Tanaka This is my paper
classification cs.LGcond-mat.stat-mech
keywords RNAinversefoldingfactorizationmachinesquadraticunconstrainedbinaryoptimizationdomain-wallencodingone-hotnormalizedensembledefectblack-boxnucleotide-to-integerassignment
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

FMQA — a surrogate-based black-box optimizer in which a factorization machine is trained on evaluated samples and then solved as a quadratic binary problem — is applied here to the RNA inverse folding problem, where the goal is to find a nucleotide sequence that folds into a given secondary structure and the cost of each sequence evaluation is high. The paper claims that the choice of binary encoding for the four nucleotides is a first-order factor: one-hot and domain-wall encodings produce much lower normalized ensemble defect and higher success rates than binary or unary encodings, and this ordering is stable across hyperparameter settings. It then shows a more subtle effect: domain-wall encoding visits its two boundary integer states (0 and 3) more often than the middle states, so which nucleotide sits at the boundary matters. Assigning G and C to the boundaries enriches GC pairs in stem regions, lowers the minimum free energy of the designed sequences, and yields lower ensemble defect than one-hot encoding. A sympathetic reader would take away that encoding design is not bookkeeping but part of the optimization strategy, and that FMQA is an evaluation-efficient starting point for RNA design.

What carries the argument

The load-bearing device is the exact correspondence between the factorization machine (FM) surrogate and a quadratic unconstrained binary optimization (QUBO) model: the FM's pairwise interaction terms become the off-diagonal QUBO entries and its linear terms the diagonal, so each new FM training round can be followed by a direct QUBO solve on an annealing-based Ising machine. The second mechanism is the encoding-dependent search landscape: domain-wall encoding represents each integer by a run of 1s followed by 0s, so the feasible subspace has a linear chain in which the boundary integers 0 and 3 have only one neighbor while 1 and 2 have two, biasing transitions and making boundary states sti

What would settle it

Take the success sequences produced by one-hot and by domain-wall (G,A,U,C) and determine their actual folds experimentally — e.g., with an in vitro structure-probing assay or by measuring melting temperatures. If domain-wall sequences do not adopt the target structure at least as often as one-hot ones, the main claim fails. A cheaper computational check: re-run the identical FMQA pipeline with a different folding-energy parameter set; if the ordering of encodings reverses or one-hot regains parity, the assignment-dependent result is an artifact of that specific in-silico oracle.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is empirical: when FMQA is used to minimize normalized ensemble defect on RNA target structures, the four standard binary-integer encodings are not interchangeable. One-hot and domain-wall encodings consistently outperform binary and unary encodings in final NED and success rate across all 24 nucleotide-to-integer assignments, while one-hot remains robust to assignment and domain-wall's outcome depends heavily on it. The mechanism the authors identify is a search bias inherent to domain-wall encoding: its feasible states form a chain in which boundary integers have only one neighbor and are therefore harder to leave, so they appear more frequen

Load-bearing premise

The load-bearing assumption is that the in-silico thermodynamic model used to compute normalized ensemble defect and minimum free energy (nearest-neighbor parameters at 37 °C with dangles treatment) faithfully predicts which sequences really fold into the target structure; if that model is wrong, the encoding ranking and the G/C-boundary advantage are claims about a simulator, not about RNA.

Editorial extensions

If this is right

  • Users of FMQA on categorical design problems should treat encoding as a hyperparameter; one-hot and domain-wall are safer than binary/unary, and the reported ordering holds for every K tested.
  • Domain-wall encoding gives designers a free dial: the nucleotide assigned to integers 0 and 3 is expressed more often, so stability-relevant symbols can be placed there to shape the solution distribution.
  • On short benchmark structures FMQA reaches lower NED with fewer evaluations than TPE, GA, and random search under identical initial data and budget, which is the regime that matters when wet-lab evaluation is the bottleneck.
  • FMQA's success follows the known difficulty of the target structure: short unstable stems (two base pairs) remain failures, while longer bulged stems are handled fine; the method inherits the field's difficulty landscape rather than removing it.

Reading between the lines

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

  • The boundary-stickiness of domain-wall encoding is a geometric property of the encoding, not of RNA; the same trick of placing high-value categories at boundary states should transfer to other FMQA categorical optimizations (e.g., protein or small-molecule design), though the paper only demonstrates it for RNA.
  • The paper reports a computational benchmark; whether the designed sequences fold correctly in a test tube is untested. A direct extension is to synthesize the FMQA-optimized sequences and check secondary structure by probing or melting experiments — that comparison would also test the assumption that the thermodynamic model's MFE ranking is physiologically meaningful.
  • The explanation for unary encoding's failure (representational degeneracy confusing the FM surrogate) predicts that any redundant encoding will degrade FMQA; an easy test is to measure FM training loss per encoding and see whether the degenerate encoding indeed has the worst fit.
  • Since the paper uses only a few structures per length and all come from the same benchmark family, scaling behavior on arbitrary pseudoknotted or long structures remains open; a natural next experiment is a length sweep with fixed encoding and assignment.
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

3 major / 4 minor

Summary. The paper proposes using factorization machine with quadratic-optimization annealing (FMQA) for RNA inverse folding, taking the normalized ensemble defect (NED) as the black-box objective. It evaluates four binary-integer encodings (one-hot, domain-wall, binary, unary) and all 24 integer-to-nucleotide assignments on the Eterna100 'stickshift' target, then compares FMQA to TPE, a genetic algorithm, and random search in terms of objective-function evaluations, and tests one configuration on eight additional Eterna100 structures. The central claims are that one-hot and domain-wall encodings outperform binary and unary encodings, that domain-wall encoding overrepresents boundary integers and can be exploited by placing G/C at the boundaries to improve thermodynamic stability, and that FMQA achieves lower NED with fewer evaluations than the baselines.

Significance. The systematic enumeration of all 24 assignments across four encodings with 10 independent runs is a useful empirical contribution, and the encoding ranking is consistent across the K sweep (Fig. 10) and most of the additional structures (Fig. 8). The boundary-assignment observation is novel and potentially actionable. However, the paper does not provide code or data availability, and the headline efficiency comparison has a selection-cost flaw that affects the paper's strongest claim; the assignment-generalization claim is based on a single target. With those points addressed, the work would be a solid empirical contribution to discrete black-box optimization for RNA design.

major comments (3)
  1. [Results—'Comparison of the number of BB function evaluations' (Fig. 6); Methods—'FMQA settings'] The efficiency claim 'FMQA achieved lower NED values with fewer evaluations' is not supported as stated because the FMQA configuration was selected on the same target used in Fig. 6. The assignment (G,A,U,C) was chosen because it gave the best domain-wall NED in the preceding 24-assignment sweep on stickshift, and K=12 was chosen from the K sweep on stickshift. That selection process consumes on the order of 10^5–10^6 NED evaluations (the 24-assignment sweep alone is 24 assignments × 10 runs × 1010 evaluations; the K sweep adds 4 K values × 4 encodings × 10 runs × 1010 evaluations) before the comparison begins. Under the paper's cost model, where each evaluation may be an expensive wet-lab experiment, this tuning cost must be included, or a held-out selection rule must be used; otherwise the comparison is biased in favor of FMQA by construction.
  2. [Abstract; Results—'Evaluation of effect...' (Figs. 3–5, 9); Discussion] The abstract states that assigning guanine and cytosine to the boundary integers 'promoted their enrichment in stem regions, which led to more thermodynamically stable secondary structures than those obtained with one-hot encoding.' This conclusion is derived from a single target (stickshift). The additional eight structures (Fig. 8) are tested only with the (G,A,U,C) assignment, so they cannot separate the assignment effect from the encoding effect. Without comparing at least one non-G/C-boundary assignment on additional targets, this statement is overgeneralized. Please either add such experiments or present the boundary-assignment result as a target-specific observation rather than a general property.
  3. [Results—'Performance on multiple target secondary structures' (Fig. 7g); Methods—'RNA secondary structure analysis'] The target named 'Prion Pseudoknot – Difficulty Level 0' appears to contain a pseudoknot, yet the paper states that pseudoknotted structures are not supported by the folding model and that only pseudoknot-free structures are considered. If this target is genuinely pseudoknotted, the NED and MFE-success metrics used here are not well-defined for it, because the target structure cannot belong to the pseudoknot-free ensemble. Please clarify whether this target is actually pseudoknot-free despite its name, or remove it and adjust the corresponding discussion.
minor comments (4)
  1. [Figs. 3, 4, 8 captions] The captions say the whiskers denote the maximum and minimum NED values, yet outliers are also plotted. Clarify whether the whiskers are 1.5×IQR or data extremes.
  2. [Fig. 6; Methods—'Baseline optimization methods'] The x-axis is labeled 'The number of BB function evaluations,' but the Methods state that the GA budget excludes evaluations of the initial dataset. Specify whether the plotted count includes the initial 10 evaluations; otherwise the comparison is ambiguous.
  3. [Methods—'Binary-integer encoding method'] For unary encoding, it would be clearer to state explicitly that every 3-bit string maps to an integer by its Hamming weight, so there are no infeasible strings; the current phrasing 'no penalty term is required' is terse.
  4. [General] No code or data availability statement is provided. For a benchmark study, making the implementation available would substantially improve reproducibility.

Circularity Check

1 steps flagged · score 6.0 of 10

Efficiency claim 'fewer evaluations' is an artifact of excluding target-specific selection of nucleotide assignment and K on the same benchmark target.

  1. fitted input called prediction [Comparison of the number of BB function evaluations; Methods: FMQA settings]
    "For FMQA, we used the integer-to-nucleotide assignment (G, A, U, C), which achieved a success rate of 1.0, low MFE values, and the lowest NED among the domain-wall encoding results in the previous subsection. ... We evaluated K=4,8,12, and 16 ... based on these results ... we selected K=12 for the FMQA experiments reported in the Results section. ... As shown in Fig. 6, FMQA achieved lower NED values with fewer evaluations compared to the other methods."

    The FMQA configuration compared in Fig. 6 was selected using NED evaluations on the same target (stickshift): the previous subsection ran 24 assignments × 10 runs with 1010 evaluations each, and the Methods K-sweep ran 4 K values × 10 runs × 1010 evaluations. These ~280,000 target-specific evaluations are not counted in Fig. 6, whose axis begins at 0 and counts only 10 initial plus 1000 FMQA iterations. The baselines (TPE, GA, RS) received no equivalent target-specific tuning, so the claim that FMQA needs 'fewer evaluations' is not an independent prediction but an artifact of pre-fitting the assignment and K to the same test instance and then omitting the fitting cost from the counter.

full rationale

The paper is primarily an empirical benchmark, not a derivation, and most of its content is self-contained. The encoding ranking (one-hot/domain-wall vs binary/unary) and the assignment-boundary effect are demonstrated directly in Figs. 3–5, 8–9; they do not reduce to a fitted input or to a self-citation. Self-citations (refs. 61, 66) support context and interpretation, but the central observations are reproduced in this paper's own experiments, so they are not load-bearing circularity. The one substantive circular step is the efficiency comparison in Fig. 6: the best integer-to-nucleotide assignment and the FM hyperparameter K are chosen by exhaustive evaluation on the same target structure, then the paper reports 'FMQA achieved lower NED values with fewer evaluations' using an evaluation counter that excludes those tuning evaluations. This is a fitted configuration being reported as a prediction on the same data it was fitted to, which makes the headline query-efficiency claim partly circular. The other claims (encoding quality, assignment bias, thermodynamic interpretation) retain independent empirical content, so the overall score is moderate rather than maximal.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claims rest on: (1) ViennaRNA's energy model as an accepted in-silico oracle; (2) the FM-QUBO equivalence solved by a stochastic SA-based Ising machine; (3) target/benchmark selection from Eterna100; and (4) configuration choices (K=12, mu=2, M=10, 1000 iterations, 2000 ms solver budget). These are not introduced as evidence of a new law; they are the experimental setup. No new physical entities are postulated.

free parameters (6)
  • FM latent dimension K = 12 (chosen from {4,8,12,16})
    Selected using stickshift with assignment (G,A,U,C); Fig. 10 shows NED/success/MFE depend on K, so the chosen value directly shapes the reported results.
  • Penalty coefficient mu (one-hot/domain-wall) = 2
    Hand-set; the paper notes no infeasible solutions were observed, but does not test sensitivity, and mu alters the QUBO landscape solved by the Ising machine.
  • Initial dataset size M = 10
    Number of random sequences seeding FMQA and baselines; affects surrogate quality and the shape of all learning curves in Fig. 6.
  • FMQA iterations / evaluation budget = 1000
    Fixed budget for all methods; the 'fewer evaluations' claim is relative to this budget and the early part of the curves.
  • Ising-machine time limit per run = 2000 ms
    SA solver budget affects the quality of each proposed candidate; not swept.
  • FM training hyperparameters = lr=0.01, 1000 epochs, AdamW, MSE
    Standard settings, but the FM surrogate quality—and hence FMQA's solutions—depend on these choices; no sensitivity reported.
assumptions (5)
  • domain assumption ViennaRNA nearest-neighbor model (Turner parameters, dangles=2, 37°C) is a faithful oracle for RNA secondary structure and ensemble defect.
    All NED, success-rate and MFE results are computed with ViennaRNA 2.7.2; no wet-lab validation. If this model mis-ranks sequences, the reported stability and efficiency conclusions inherit those errors.
  • domain assumption Pseudoknot-free target structures only.
    Stated in Results; however the Eterna100 target 'Prion Pseudoknot – Difficulty Level 0' is used without clarification of whether it is pseudoknot-free.
  • domain assumption The Boltzmann-weighted ensemble defect (Eq. 5) is an appropriate objective for RNA inverse folding.
    Adopted from refs 67,68; the surrogate and all baselines minimize this quantity, so the comparison is about this objective, not about alternate design metrics.
  • standard math FM-QUBO equivalence and approximate minimization by an SA-based Ising machine.
    Uses x_i^2 = x_i for diagonal terms; the SA solver returns approximate ground states within 2000 ms. Stochasticity of the solver is a noise source in all comparisons.
  • ad hoc to paper Unary encoding's many-to-one binary strings (Table 1) can be handled by the FM without explicit constraint.
    The paper argues redundant representations complicate surrogate learning; this is presented in the Discussion as an explanation of poor unary performance, but the decoding of arbitrary binary strings is not formally specified in the FMQA loop.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Factorization Machine with Quadratic-Optimization Annealing for RNA Inverse Folding and Evaluation of Binary-Integer Encoding and Nucleotide Assignment." pith.science (2026). https://pith.science/paper/NRVBTI6A

@misc{pith2026260216643,
  author       = {Pith},
  title        = {Pith review of: Factorization Machine with Quadratic-Optimization Annealing for RNA Inverse Folding and Evaluation of Binary-Integer Encoding and Nucleotide Assignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NRVBTI6A}},
  note         = {Machine review of arXiv:2602.16643}
}
read the original abstract

The RNA inverse folding problem aims to identify nucleotide sequences that preferentially adopt a given target secondary structure. While various heuristic and machine learning-based approaches have been proposed, many require a large number of sequence evaluations, which limits their applicability when experimental validation is costly. We propose a method to solve the problem using a factorization machine with quadratic-optimization annealing (FMQA). FMQA is a discrete black-box optimization method reported to obtain high-quality solutions with a limited number of evaluations. Applying FMQA to the problem requires converting nucleotides into binary variables. However, the influence of integer-to-nucleotide assignments and binary-integer encoding on the performance of FMQA has not been thoroughly investigated, even though such choices determine the structure of the surrogate model and the search landscape, and thus can directly affect solution quality. Therefore, this study aims both to establish a novel FMQA framework for RNA inverse folding and to analyze the effects of these assignments and encoding methods. We evaluated all 24 possible assignments of the four nucleotides to the ordered integers (0-3), in combination with four binary-integer encoding methods. Our results demonstrated that one-hot and domain-wall encodings outperform binary and unary encodings in terms of the normalized ensemble defect value. In domain-wall encoding, nucleotides assigned to the boundary integers (0 and 3) appeared with higher frequency. In the RNA inverse folding problem, assigning guanine and cytosine to these boundary integers promoted their enrichment in stem regions, which led to more thermodynamically stable secondary structures than those obtained with one-hot encoding.

Figures

Figures reproduced from arXiv: 2602.16643 by the authors.

Figure 1
Figure 1. Overview of the RNA inverse folding problem. White circles represent arbitrary nucleotides in target secondary structure, and black lines indicate base pairs formed through hydrogen bonding. A, U, G, and C denote adenine, uracil, guanine, and cytosine, respectively. the target structure. Other approaches use stochastic local search18–21, genetic algorithms22–24, constraint programming25, ant colony optimization26, a… view at source ↗
Figure 2
Figure 2. Schematic illustration of the proposed FMQA for the RNA inverse folding problem. As an example, the BB function is defined as the NED. Integer Binary variable sequence One-hot encoding Domain-wall encoding Binary encoding Unary encoding 0 1000 000 00 000 1 0100 100 10 100, 010, 001 2 0010 110 01 110, 101, 011 3 0001 111 11 111 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Normalized ensemble defect values obtained by FMQA under different combinations of binary-integer encoding methods and integer-to-nucleotide assignments. Crosses indicate the average NED over 10 runs. The upper and lower whiskers denote the maximum and minimum NED values, respectively. Black circle, red triangle, blue square, and green diamond represent outliers. Panels in the top row correspond to assignments in wh… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Success rate obtained by FMQA under different combinations of binary-integer encoding methods and integer-to-nucleotide assignments. Panels in the top row correspond to assignments in which A or U was assigned to integer 0, whereas panels in the bottom row correspond t…
Figure 5
Figure 5. Figure 5: Minimum free energy values obtained by FMQA under different combinations of binary-integer encoding methods and integer-to-nucleotide assignments. Only success solutions are plotted. When two or more success solutions were obtained, their average MFE value is indicated…
Figure 6
Figure 6. Figure 6: Relationship between the number of BB function evaluations and the best normalized ensemble defect value up to each evaluation [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Target secondary structures selected from the Eterna100 benchmark and evaluated in this study. White circles represent nucleotides, and black lines indicate base pairs formed by hydrogen bonding. The target structures are: (a) G-C Placement (12 nt), (b) Simple Hairpin …
Figure 8
Figure 8. Figure 8: Performance metrics on multiple Eterna100 benchmark target structures. (a) Normalized ensemble defect values obtained by FMQA for each target structure. Crosses indicate the average NED over 10 independent runs. The upper and lower whiskers denote the maximum and minim…
Figure 9
Figure 9. Figure 9: Nucleotide frequencies obtained from the success solutions across different integer-to-nucleotide assignments. (a) Nucleotide frequency in stem regions, (b) nucleotide frequency in non-stem regions, and (c) nucleotide frequency over the entire RNA sequence. The baselin…
Figure 10
Figure 10. Figure 10: Performance metrics for different values of the hyperparameter K under each binary-integer encoding method. (a) Normalized ensemble defect values obtained by FMQA for each encoding method and value of K. Crosses indicate the average NED over 10 independent runs. The u…

Discussion (0). Sign in 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. Stage-dependent integer-binary encoding in factorization-machine black-box optimization

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    Stage-dependent encoding in FMQA black-box optimization, using one-hot for learning and domain-wall for search, improves residual error on discretized Rastrigin functions under finer discretization and higher dimensio...

Reference graph

Works this paper leans on

47 extracted references · 6 linked inside Pith · cited by 1 Pith paper

  1. [4]

    & Patel, D

    Serganov, A. & Patel, D. J. Ribozymes, riboswitches and beyond: regulation of gene expression without proteins.Nat. Rev, Genet.8, 776–790 (2007)

    2007

  2. [5]

    J., Porter, F

    Pardi, N., Hogan, M. J., Porter, F. W. & Weissman, D. mRNA vaccines—a new era in vaccinology.Nat. Rev. Drug Discov. 17, 261–279 (2018). 6.Hamada, M. In silico approaches to RNA aptamer design.Biochimie145, 8–14 (2018)

    2018

  3. [7]

    & Dhar, P

    Singh, V ., Braddick, D. & Dhar, P. K. Exploring the potential of genome editing CRISPR-Cas9 technology.Gene599, 1–18 (2017)

    2017

  4. [8]

    Jaffrey, S. R. RNA-based fluorescent biosensors for detecting metabolites in vitro and in living cells.Adv. Pharmacol.82, 187–203 (2018). 11/17 9.Bauer, G. & Suess, B. Engineered riboswitches as novel tools in molecular biology.J. Biotechnol.124, 4–11 (2006). 10.Dixon, N.et al.Reengineering orthogonally selective riboswitches.Proc. Natl. Acad. Sci.107, 28...

    2018

  5. [11]

    C., Rosenberg, J

    Seeman, N. C., Rosenberg, J. M., Suddath, F., Kim, J. J. P. & Rich, A. RNA double-helical fragments at atomic resolution: I. The crystal and molecular structure of sodium adenylyl-3’, 5’-uridine hexahydrate.J. Mol. Biol.104, 109–144 (1976)

    1976

  6. [12]

    M., Seeman, N

    Rosenberg, J. M., Seeman, N. C., Day, R. O. & Rich, A. RNA double-helical fragments at atomic resolution: II. the crystal structure of sodium guanylyl-3’, 5’-cytidine nonahydrate.J. Mol. Biol.104, 145–167 (1976). 13.Varani, G. & McClain, W. H. The G· U wobble base pair.EMBO Rep.1, 18–23 (2000). 14.Reese, C. B. Oligo-and poly-nucleotides: 50 years of chemi...

    1976

  7. [16]

    & Berger, B

    Schnall-Levin, M., Chindelevitch, L. & Berger, B. Inverting the Viterbi algorithm: an abstract framework for structure design. InProceedings of the 25th international conference on machine learning, 904–911 (2008)

    2008

  8. [17]

    & Sikora, F

    Bonnet, É., Rz ˛ a˙zewski, P. & Sikora, F. Designing RNA secondary structures is hard.J. Comput. Biol.27, 302–316 (2020)

    2020

Show all 47 references
  1. [18]

    P., Hutter, F., Hoos, H

    Andronescu, M., Fejes, A. P., Hutter, F., Hoos, H. H. & Condon, A. A new algorithm for RNA secondary structure design. J. Mol. Biol.336, 607–624 (2004). 19.Busch, A. & Backofen, R. INFO-RNA—a fast approach to inverse RNA folding.Bioinformatics22, 1823–1831 (2006)

    2004

  2. [20]

    N., Wolfe, B

    Zadeh, J. N., Wolfe, B. R. & Pierce, N. A. Nucleic acid sequence design via efficient ensemble defect optimization.J. Comput. Chem.32, 439–452 (2011). 21.Zadeh, J. N.et al.NUPACK: Analysis and design of nucleic acid systems.J. Comput. Chem.32, 170–173 (2011). 22.Taneda, A. MOD...

    2011

  3. [24]

    & Vega-Rodríguez, M

    Rubio-Largo, Á., Vanneschi, L., Castelli, M. & Vega-Rodríguez, M. A. Multiobjective metaheuristic to design RNA sequences.IEEE Trans. Evol. Comput.23, 156–169 (2019). 25.Garcia-Martin, J. A., Clote, P. & Dotu, I. RNAiFOLD: a constraint programming algorithm for RNA inverse fol...

    2019

  4. [26]

    & Backofen, R

    Kleinkauf, R., Mann, M. & Backofen, R. antaRNA: ant colony-based RNA sequence design.Bioinformatics31, 3114–3121 (2015)

    2015

  5. [27]

    & Tsuda, K

    Yang, X., Yoshizoe, K., Taneda, A. & Tsuda, K. RNA inverse folding using Monte Carlo tree search.BMC Bioinforma.18, 468 (2017)

    2017

  6. [28]

    & Pande, V

    Shi, J., Das, R. & Pande, V . S. SentRNA: Improving computational RNA design by incorporating a prior of human design strategies.arXiv preprint arXiv:1803.03146(2018)

    2018 arXiv

  7. [29]

    V .et al.EternaBrain: automated RNA design through move sets and strategies from an internet-scale RNA videogame.PLOS Comput

    Koodli, R. V .et al.EternaBrain: automated RNA design through move sets and strategies from an internet-scale RNA videogame.PLOS Comput. Biol.15, e1007059 (2019)

    2019

  8. [30]

    & Pande, V

    Eastman, P., Shi, J., Ramsundar, B. & Pande, V . S. Solving the RNA design problem with reinforcement learning.PLOS Comput. Biol.14, e1006176 (2018). 31.Runge, F., Stoll, D., Falkner, S. & Hutter, F. Learning to design RNA.arXiv preprint arXiv:1812.11951(2018). 32.Sumi, S., Ha...

    2018 arXiv

  9. [33]

    & Sakakibara, Y

    Akiyama, M. & Sakakibara, Y . Informative RNA base embedding for RNA structural alignment and clustering by deep representation learning.NAR Genom. Bioinform.4, lqac012 (2022)

    2022

  10. [34]

    R., Schonlau, M

    Jones, D. R., Schonlau, M. & Welch, W. J. Efficient global optimization of expensive black-box functions.J. Glob. Optim. 13, 455–492 (1998). 35.Forrester, A. I. & Keane, A. J. Recent advances in surrogate-based optimization.Prog. Aerosp. Sci.45, 50–79 (2009)

    1998

  11. [36]

    Multidiscip

    Ramu, P.et al.A survey of machine learning techniques in structural and multidisciplinary optimization.Struct. Multidiscip. Optim.65, 266 (2022)

    2022

  12. [37]

    38.Rendle, S

    Tamura, R.et al.Black-box optimization using factorization and ising machines.arXiv preprint arXiv:2507.18003(2025). 38.Rendle, S. Factorization machines. In2010 IEEE International conference on data mining, 995–1000 (2010). 12/17

    2025 arXiv

  13. [39]

    Kitai, K.et al.Designing metamaterials with quantum annealing and factorization machines.Phys. Rev. Res.2(2020). Art. no. 013319

    2020

  14. [40]

    Mohseni, N., McMahon, P. L. & Byrnes, T. Ising machines as hardware solvers of combinatorial optimization problems. Nat. Rev. Phys.4, 363–379 (2022). 41.Johnson, M. W.et al.Quantum annealing with manufactured spins.Nature473, 194–198 (2011). 42.Kadowaki, T. & Nishimori, H. Qua...

    2022

  15. [43]

    & Sipser, M

    Farhi, E., Goldstone, J., Gutmann, S. & Sipser, M. Quantum computation by adiabatic evolution.arXiv preprint quant-ph/0001106(2000)

    2000 arXiv

  16. [44]

    & Tamura, H

    Tsukamoto, S., Takatsu, M., Matsubara, S. & Tamura, H. An accelerator architecture for combinatorial optimization problems.Fujitsu Sci. Tech. J.53, 8–13 (2017)

    2017

  17. [45]

    & Yamaoka, M

    Okuyama, T., Hayashi, M. & Yamaoka, M. An Ising computer based on simulated quantum annealing by path integral Monte Carlo method. In2017 IEEE international conference on rebooting computing (ICRC), 1–6 (IEEE, 2017)

    2017

  18. [46]

    Adv.7(2021)

    Goto, H.et al.High-performance combinatorial optimization based on classical mechanics.Sci. Adv.7(2021). Art. no. eabe7953. 47.Fixstars Amplify. Fixstars Amplify Annealing Engine. https://amplify.fixstars.com/en/. 48.Kirkpatrick, S., Gelatt, C. D. & Vecchi, M. P. Optimization ...

    2021

  19. [49]

    S., Aragon, C

    Johnson, D. S., Aragon, C. R., McGeoch, L. A. & Schevon, C. Optimization by simulated annealing: An experimental evaluation; part II, graph coloring and number partitioning.Oper. Res.39, 378–406 (1991)

    1991

  20. [50]

    & Chakrabarti, B

    Das, A. & Chakrabarti, B. K.Quantum annealing and related optimization methods, vol. 679 (Springer Science & Business Media, 2005)

    2005

  21. [51]

    & Dixon, A

    Goto, H., Tatsumura, K. & Dixon, A. R. Combinatorial optimization by simulating adiabatic bifurcations in nonlinear Hamiltonian systems.Sci. Adv.5(2019). Art. no. eaav2372

    2019

  22. [52]

    & Miura, Y

    Nawa, K., Suzuki, T., Masuda, K., Tanaka, S. & Miura, Y . Quantum annealing optimization method for the design of barrier materials in magnetic tunnel junctions.Phys. Rev. Appl.20(2023). Art. no. 024044. 53.Kim, S.et al.Quantum annealing-aided design of an ultrathin-metamateri...

    2023

  23. [54]

    Couzinié, Y .et al.Machine learning supported annealing for prediction of grand canonical crystal structures.J. Phys. Soc. Jpn.94, 044802 (2025)

    2025

  24. [55]

    Tucs, A.et al.Quantum annealing designs nonhemolytic antimicrobial peptides in a discrete latent space.ACS Med. Chem. Lett.14, 577–582 (2023)

    2023

  25. [56]

    Huang, X. & Ju, S. Tutorial: AI-assisted exploration and active design of polymers with high intrinsic thermal conductivity. J. Appl. Phys.135(2024)

    2024

  26. [57]

    & Kadowaki, T

    Matsumori, T., Taki, M. & Kadowaki, T. Application of QUBO solver using black-box optimization to structural design for resonance avoidance.Sci. Rep.12(2022). Art. no. 12143

    2022

  27. [58]

    Express30, 43503–43512 (2022)

    Inoue, T.et al.Towards optimization of photonic-crystal surface-emitting lasers via quantum annealing.Opt. Express30, 43503–43512 (2022)

    2022

  28. [59]

    Tamura, R.et al.Machine learning prediction of the mechanical properties of injection-molded polypropylene through X-ray diffraction analysis.Sci. Technol. Adv. Mater.25, 2388016 (2024)

    2024

  29. [60]

    & Tanaka, S

    Kikuchi, S. & Tanaka, S. High-order epistasis detection using factorization machine with quadratic optimization annealing and MDR-based evaluation.arXiv preprint arXiv:2602.01860(2026)

    2026

  30. [61]

    & Tanaka, S

    Seki, Y ., Tamura, R. & Tanaka, S. Black-box optimization for integer-variable problems using Ising machines and factorization machines.arXiv preprint arXiv:2209.01016(2022)

    2022 arXiv

  31. [62]

    & Yoshida, T

    Koshikawa, S., Hosaka, A. & Yoshida, T. Efficient bit labeling in factorization machines with annealing for traveling salesman problem.Sci. Rep.15, 26910 (2025)

    2025

  32. [63]

    Domain wall encoding of discrete variables for quantum annealing and QAOA.Quantum Sci

    Chancellor, N. Domain wall encoding of discrete variables for quantum annealing and QAOA.Quantum Sci. Technol.4 (2019). Art. no. 045004

    2019

  33. [64]

    & Togawa, N

    Tamura, K., Shirai, T., Katsura, H., Tanaka, S. & Togawa, N. Performance comparison of typical binary-integer encodings in an Ising machine.IEEE Access9, 81032–81039 (2021). 13/17

    2021

  34. [65]

    & Chancellor, N

    Chen, J., Stollenwerk, T. & Chancellor, N. Performance of domain-wall encoding for quantum annealing.IEEE Trans. Quantum Eng.2, 1–14 (2021)

    2021

  35. [66]

    & Tanaka, S

    Kikuchi, S., Takahashi, K. & Tanaka, S. Performance of domain-wall encoding in digital Ising machine.arXiv preprint arXiv:2410.11198(2024)

    2024 arXiv

  36. [67]

    M., Lin, M., Winfree, E

    Dirks, R. M., Lin, M., Winfree, E. & Pierce, N. A. Paradigms for computational nucleic acid design.Nucleic Acids Res. 32, 1392–1403 (2004). 68.Ward, M., Courtney, E. & Rivas, E. Fitness functions for RNA structure design.Nucleic Acids Res.51, e40–e40 (2023)

    2004

  37. [69]

    Turner, D. H. & Mathews, D. H. NNDB: the nearest neighbor parameter database for predicting stability of nucleic acid secondary structure.Nucleic Acids Res.38, D280–D282 (2010). 70.Lorenz, R.et al.ViennaRNA package 2.0.Algorithms Mol. Biol.6, 1–14 (2011)

    2010

  38. [71]

    Anderson-Lee, J.et al.Principles for predicting RNA secondary structure design difficulty.J. Mol. Biol.428, 748–757 (2016)

    2016

  39. [72]

    Promoting the application of advanced quantum technology platforms to social issues

    Mathews, D. H.et al.Incorporating chemical modification constraints into a dynamic programming algorithm for prediction of RNA secondary structure.Proc. Natl. Acad. Sci.101, 7287–7292 (2004). 73.Blank, J. & Deb, K. pymoo: Multi-objective optimization in Python.IEEE Access8, 89...

    2004

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.