pith. sign in

arxiv: 2606.07562 · v1 · pith:TD6RABR5new · submitted 2026-05-25 · 🧬 q-bio.BM · cs.AI

The Montparnasse Algorithm for RNA Design

Tristan Cazenave This is my paper

Pith reviewed 2026-06-29 19:30 UTC · model grok-4.3

classification 🧬 q-bio.BM cs.AI
keywords RNA designMonte Carlo searchsecondary structureEterna100nucleotide sequencepolicy adaptationmulticriteria evaluationmessenger RNA
0
0 comments X

The pith

Montparnasse solves RNA design puzzles more than three times faster than previous methods.

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

The paper presents Montparnasse, a Monte Carlo search framework for RNA design that optimizes nucleotide sequences for given secondary structures. It augments Generalized Nested Rollout Policy Adaptation with a problem-specific prior, slow adaptation, and lexicographic multicriteria evaluation. This allows it to solve all 100 puzzles in the Eterna100 V1 benchmark consistently faster than the prior state of the art across different time limits. It also identifies better sequences for hemoglobin alpha mRNA than standard optimization methods. If correct, this would mean faster and more effective tools for creating designed RNAs in biology and medicine.

Core claim

Montparnasse is a Monte Carlo search framework based on Generalized Nested Rollout Policy Adaptation, augmented with a problem-specific prior, slow and long adaptation at level 1, and a lexicographic multicriteria evaluation. It solves all 100 puzzles of the Eterna100 V1 benchmark consistently faster than DesiRNA across all time limits, reaching full coverage more than three times faster overall. On messenger RNA secondary structure optimization for hemoglobin alpha, it identifies sequences with more paired bases than the MFE-optimal solution of LinearDesign.

What carries the argument

Generalized Nested Rollout Policy Adaptation augmented with a problem-specific prior, slow and long adaptation at level 1, and lexicographic multicriteria evaluation that guides the search for optimal RNA sequences.

If this is right

  • The algorithm achieves full coverage of the Eterna100 V1 benchmark more than three times faster than previous methods.
  • It consistently outperforms DesiRNA across all tested time limits on the benchmark puzzles.
  • It produces sequences with more paired bases than the minimum free energy optimal solution for hemoglobin alpha mRNA.
  • The method applies to problems in synthetic biology, medicine, and nanotechnology.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the approach works on other RNA design tasks, it could speed up the creation of custom RNA molecules for medical uses.
  • The multicriteria evaluation might help rank designs better for real-world applications beyond the benchmark.
  • Adapting the framework to larger or more complex RNA structures could test its scalability.
  • Combining it with experimental validation would show if the in silico improvements translate to better biological function.

Load-bearing premise

That success on the Eterna100 V1 benchmark and the single hemoglobin example means the method will work well on the RNA design problems that matter most in practice.

What would settle it

A new collection of RNA design problems where Montparnasse does not reach solutions faster than DesiRNA or where the resulting sequences fail to meet practical performance criteria in experiments.

Figures

Figures reproduced from arXiv: 2606.07562 by Tristan Cazenave.

Figure 1
Figure 1. Figure 1: Evolution of the average base-pair distance (BPD) on problem 90 as a function of the number of folding evaluations [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: mRNA secondary structure optimization for hemoglobin alpha (76 amino acids, 228 nucleotides) over 600 seconds. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Gladius: target secondary structure of Eterna100 [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
read the original abstract

RNA design consists of discovering a nucleotide sequence that optimizes predefined criteria, such as secondary structure. It is useful for synthetic biology, medicine, and nanotechnology. We propose Montparnasse, a Monte Carlo search framework based on Generalized Nested Rollout Policy Adaptation, augmented with a problem-specific prior, slow and long adaptation at level 1, and a lexicographic multicriteria evaluation. Montparnasse solves all 100 puzzles of the Eterna100 V1 benchmark consistently faster than DesiRNA, the previous state of the art, across all time limits, reaching full coverage more than three times faster overall. On messenger RNA secondary structure optimization for hemoglobin alpha, it identifies sequences with more paired bases than the MFE-optimal solution of LinearDesign.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper introduces Montparnasse, a Monte Carlo search framework based on Generalized Nested Rollout Policy Adaptation augmented with a problem-specific prior, slow/long adaptation at level 1, and lexicographic multicriteria evaluation. It claims that this algorithm solves all 100 puzzles of the Eterna100 V1 benchmark consistently faster than DesiRNA across all time limits (reaching full coverage more than three times faster overall) and identifies mRNA sequences for hemoglobin alpha with more paired bases than the MFE-optimal solution from LinearDesign.

Significance. If the performance claims hold under rigorous statistical validation, the work would constitute a meaningful advance in computational RNA design by improving both speed and solution quality on a standard benchmark, with direct relevance to synthetic biology and mRNA therapeutics. The explicit use of the Eterna100 V1 benchmark and a multicriteria evaluation scheme are strengths that facilitate comparison and downstream applicability.

major comments (1)
  1. [Results (performance claims on Eterna100 V1)] The central claim that Montparnasse solves the Eterna100 V1 puzzles 'consistently faster' than DesiRNA rests on an unverified assumption for a stochastic Monte Carlo method; the manuscript provides no evidence of multiple independent runs per puzzle, reported means/variances, or statistical tests (e.g., Wilcoxon or t-tests on runtimes), so single-trial times cannot support the 'consistently' qualifier or the 'more than three times faster overall' aggregate.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful review and constructive feedback on our manuscript. We address the major comment regarding the statistical validation of our performance claims point by point below.

read point-by-point responses

  1. Referee: [Results (performance claims on Eterna100 V1)] The central claim that Montparnasse solves the Eterna100 V1 puzzles 'consistently faster' than DesiRNA rests on an unverified assumption for a stochastic Monte Carlo method; the manuscript provides no evidence of multiple independent runs per puzzle, reported means/variances, or statistical tests (e.g., Wilcoxon or t-tests on runtimes), so single-trial times cannot support the 'consistently' qualifier or the 'more than three times faster overall' aggregate.

    Authors: We agree with the referee that the current presentation of results is insufficient to support the 'consistently faster' claim for a stochastic algorithm. The manuscript reports single-run times per puzzle without multiple independent trials, means, variances, or statistical comparisons. In the revised manuscript we will add results from at least 10 independent runs per puzzle, report means and standard deviations for runtimes, and include paired statistical tests (Wilcoxon signed-rank) between Montparnasse and DesiRNA. The abstract, results section, and any aggregate claims (including the 'more than three times faster' statement) will be updated to reflect only those findings that survive this analysis. We view this as a necessary strengthening of the paper. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical performance claims on fixed benchmarks

full rationale

The paper introduces Montparnasse as a Monte Carlo search algorithm with specific augmentations and reports its runtime performance on the fixed Eterna100 V1 benchmark set and one mRNA example. These are direct empirical measurements of an implemented procedure against external test cases, with no equations, fitted parameters, self-citations, or derivations that reduce the central claims to the inputs by construction. The lexicographic evaluation and prior are presented as design choices, not as outputs derived from the results themselves.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only abstract available; ledger is therefore minimal and based solely on stated assumptions in the abstract.

axioms (1)
  • domain assumption The Eterna100 V1 benchmark and hemoglobin alpha mRNA constitute sufficient evidence for superiority claims.
    The abstract uses these two test cases to assert consistent outperformance.

pith-pipeline@v0.9.1-grok · 5643 in / 1174 out tokens · 32435 ms · 2026-06-29T19:30:59.422971+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

274 extracted references · 26 canonical work pages

  1. [1]

    Nucleic Acids Research , volume=

    DesiRNA: structure-based design of RNA sequences with a replica exchange Monte Carlo approach , author=. Nucleic Acids Research , volume=. 2025 , publisher=

    2025

  2. [2]

    Nature , volume=

    Algorithm for optimized mRNA design improves stability and immunogenicity , author=. Nature , volume=. 2023 , publisher=

    2023

  3. [3]

    Proceedings of the AAAI Conference on Artificial Intelligence , pages=

    Warm-starting nested rollout policy adaptation with optimal stopping , author=. Proceedings of the AAAI Conference on Artificial Intelligence , pages=

  4. [4]

    2022 , publisher=

    Merleau, Nono SC and Smerlak, Matteo , journal=. 2022 , publisher=

    2022

  5. [5]

    Multi-objective optimization for

    Taneda, Akito , journal=. Multi-objective optimization for. 2015 , publisher=

    2015

  6. [6]

    2006 , publisher=

    Busch, Anke and Backofen, Rolf , journal=. 2006 , publisher=

    2006

  7. [7]

    A new algorithm for

    Andronescu, Mirela and Fejes, Anthony P and Hutter, Frank and Hoos, Holger H and Condon, Anne , journal=. A new algorithm for. 2004 , publisher=

    2004

  8. [8]

    Fast folding and comparison of

    Hofacker, Ivo L and Fontana, Walter and Stadler, Peter F and Bonhoeffer, L Sebastian and Tacker, Manfred and Schuster, Peter and others , journal=. Fast folding and comparison of. 1994 , publisher=

    1994

  9. [9]

    European Workshop on Reinforcement Learning , year=

    Generalized nested rollout policy adaptation with limited repetitions , author=. European Workshop on Reinforcement Learning , year=

  10. [10]

    International Conference on Parallel Problem Solving from Nature , pages=

    Learning a Prior for Monte Carlo Search by Replaying Solutions to Combinatorial Problems , author=. International Conference on Parallel Problem Solving from Nature , pages=. 2024 , organization=

    2024

  11. [11]

    2024 , publisher=

    Cazenave, Tristan and Touzani, Hamza , booktitle=. 2024 , publisher=

    2024

  12. [12]

    2024 , publisher=

    RNA Design , author=. 2024 , publisher=

    2024

  13. [13]

    Designing

    Obonyo, Stephen and Jouandeau, Nicolas and Owuor, Dickson , booktitle=. Designing

  14. [14]

    Self-playing

    Obonyo, Stephen and Jouandeau, Nicolas and Owuor, Dickson , journal=. Self-playing. 2024 , publisher=

    2024

  15. [15]

    2024 , organization=

    Obonyo, Stephen and Jouandeau, Nicolas and Owuor, Dickson , booktitle=. 2024 , organization=

    2024

  16. [16]

    BMC bioinformatics , volume=

    RNA inverse folding using Monte Carlo tree search , author=. BMC bioinformatics , volume=. 2017 , publisher=

    2017

  17. [17]

    A Simple yet Effective Greedy Evolutionary Strategy for

    Lozano-Garc. A Simple yet Effective Greedy Evolutionary Strategy for. IEEE Transactions on Evolutionary Computation , year=

  18. [18]

    Oligo- and poly-nucleotides: 50 years of chemical synthesis

    Reese, Colin B. Oligo- and poly-nucleotides: 50 years of chemical synthesis. Org. Biomol. Chem. 2005

    2005

  19. [19]

    2016 , author =

    Principles for Predicting RNA Secondary Structure Design Difficulty , journal =. 2016 , author =

    2016

  20. [20]

    Algorithms for molecular biology , volume=

    ViennaRNA Package 2.0 , author=. Algorithms for molecular biology , volume=. 2011 , publisher=

    2011

  21. [21]

    Advances in neural information processing systems , volume=

    Attention is all you need , author=. Advances in neural information processing systems , volume=

  22. [22]

    Nucleic Acids Research , volume=

    Rfam 14: expanded coverage of metagenomic, viral and microRNA families , author=. Nucleic Acids Research , volume=. 2021 , publisher=

    2021

  23. [23]

    Handbook of metaheuristics , pages=

    Iterated local search: Framework and applications , author=. Handbook of metaheuristics , pages=. 2019 , publisher=

    2019

  24. [24]

    Advances in Computer Games: 16th International Conference, ACG 2019, Macao, China, August 11--13, 2019, Revised Selected Papers 16 , pages=

    A practical introduction to the Ludii general game system , author=. Advances in Computer Games: 16th International Conference, ACG 2019, Macao, China, August 11--13, 2019, Revised Selected Papers 16 , pages=. 2020 , organization=

    2019

  25. [25]

    International Conference on Learning Representations , year=

    Policy improvement by planning with Gumbel , author=. International Conference on Learning Representations , year=

  26. [26]

    Artificial Intelligence Review , volume=

    Monte Carlo tree search: A review of recent modifications and applications , author=. Artificial Intelligence Review , volume=. 2023 , publisher=

    2023

  27. [27]

    Advances in Neural Information Processing Systems , volume=

    Symbolic discovery of optimization algorithms , author=. Advances in Neural Information Processing Systems , volume=

  28. [28]

    Jacques Pitrat , title =

  29. [29]

    Tristan Cazenave , title =

  30. [30]

    Proceedings of the 26th annual international conference on machine learning , pages=

    Curriculum learning , author=. Proceedings of the 26th annual international conference on machine learning , pages=

  31. [31]

    Cosine Annealing, Mixnet and Swish Activation for Computer

    Cazenave, Tristan and Sentuc, Julien and Videau, Mathurin , booktitle=. Cosine Annealing, Mixnet and Swish Activation for Computer. 2021 , publisher=

    2021

  32. [32]

    Vision Transformers for Computer

    Sagri, Amani and Arjonilla, Jerome and Saffidine, Abdallah and Cazenave, Tristan , booktitle=. Vision Transformers for Computer. 2024 , publisher=

    2024

  33. [33]

    Advances in Computer Games , pages=

    Sequential halving using scores , author=. Advances in Computer Games , pages=. 2021 , publisher=

    2021

  34. [34]

    Generalized Rapid Action Value Estimation , booktitle =

    Tristan Cazenave , editor =. Generalized Rapid Action Value Estimation , booktitle =. 2015 , pages =

    2015

  35. [35]

    Quantum Machine Learning at ECML PKDD , year=

    Monte Carlo Qubit Routing , author=. Quantum Machine Learning at ECML PKDD , year=

  36. [36]

    Physical Review A , volume=

    Monte Carlo graph search for quantum circuit optimization , author=. Physical Review A , volume=. 2023 , publisher=

    2023

  37. [37]

    IEEE Transactions on Quantum Engineering , year=

    Automated quantum circuit design with nested monte carlo tree search , author=. IEEE Transactions on Quantum Engineering , year=

  38. [38]

    arXiv preprint arXiv:2402.14396 , year=

    Quantum Circuit Optimization with AlphaTensor , author=. arXiv preprint arXiv:2402.14396 , year=

    work page arXiv

  39. [39]

    Nature , volume=

    Faster sorting algorithms discovered using deep reinforcement learning , author=. Nature , volume=. 2023 , publisher=

    2023

  40. [40]

    Search Based Software Engineering: 8th International Symposium, SSBSE 2016, Raleigh, NC, USA, October 8-10, 2016, Proceedings 8 , pages=

    Field report: Applying monte carlo tree search for program synthesis , author=. Search Based Software Engineering: 8th International Symposium, SSBSE 2016, Raleigh, NC, USA, October 8-10, 2016, Proceedings 8 , pages=. 2016 , organization=

    2016

  41. [41]

    Evolving Monte Carlo tree search algorithms , author=. Dept. Inf., Univ. Paris , volume=. 2007 , publisher=

    2007

  42. [42]

    Discovering Mathematical Formulas from Data via

    Li, Yanjie and Li, Weijun and Yu, Lina and Wu, Min and Liu, Jingyi and Li, Wenqiang and Hao, Meilan and Wei, Shu and Deng, Yusong , journal=. Discovering Mathematical Formulas from Data via

  43. [43]

    2017 IEEE 29th International Conference on Tools with Artificial Intelligence (ICTAI) , pages=

    Algorithm Discovery with Monte-Carlo Search: Controlling the Size , author=. 2017 IEEE 29th International Conference on Tools with Artificial Intelligence (ICTAI) , pages=. 2017 , organization=

    2017

  44. [44]

    IEEE Transactions on Computational Intelligence and AI in Games , volume=

    Monte carlo search algorithm discovery for single-player games , author=. IEEE Transactions on Computational Intelligence and AI in Games , volume=. 2013 , publisher=

    2013

  45. [45]

    Neural Computing and Applications , volume=

    Incorporating Actor-Critic in Monte Carlo tree search for symbolic regression , author=. Neural Computing and Applications , volume=. 2021 , publisher=

    2021

  46. [46]

    Nested monte carlo search expression discovery for the automated design of fuzzy

    Illetskova, Marketa and Elnabarawy, Islam and Da Silva, Leonardo Enzo Brito and Tauritz, Daniel R and Wunsch, Donald C , booktitle=. Nested monte carlo search expression discovery for the automated design of fuzzy

  47. [47]

    2015 IEEE Symposium Series on Computational Intelligence , pages=

    Forecasting financial volatility using nested Monte Carlo expression discovery , author=. 2015 IEEE Symposium Series on Computational Intelligence , pages=. 2015 , organization=

    2015

  48. [48]

    arXiv preprint arXiv:2205.13134 , year=

    Symbolic physics learner: Discovering governing equations via monte carlo tree search , author=. arXiv preprint arXiv:2205.13134 , year=

    work page arXiv

  49. [49]

    ECAI 2010 , pages=

    Nested monte-carlo expression discovery , author=. ECAI 2010 , pages=. 2010 , publisher=

    2010

  50. [50]

    International Journal on Artificial Intelligence Tools , volume=

    Monte-carlo expression discovery , author=. International Journal on Artificial Intelligence Tools , volume=. 2013 , publisher=

    2013

  51. [51]

    Nature , volume=

    Discovering faster matrix multiplication algorithms with reinforcement learning , author=. Nature , volume=. 2022 , publisher=

    2022

  52. [52]

    2018 6th International Conference on Control Engineering & Information Technology (CEIT) , pages=

    Performance analysis and CPU vs GPU comparison for deep learning , author=. 2018 6th International Conference on Control Engineering & Information Technology (CEIT) , pages=. 2018 , organization=

    2018

  53. [53]

    Neural Computing and Applications , volume=

    CPU versus GPU: which can perform matrix computation faster—performance comparison for basic linear algebra subprograms , author=. Neural Computing and Applications , volume=. 2019 , publisher=

    2019

  54. [54]

    GPU computing in discrete optimization

    Brodtkorb, André and Hagen, Trond and Schulz, Christian and Hasle, Geir , year =. GPU computing in discrete optimization. Part I: Introduction to the GPU , volume =. EURO Journal on Transportation and Logistics , doi =

  55. [55]

    Science , volume=

    A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play , author=. Science , volume=. 2018 , publisher=

    2018

  56. [56]

    Nature , volume=

    Mastering atari, go, chess and shogi by planning with a learned model , author=. Nature , volume=. 2020 , publisher=

    2020

  57. [57]

    M. Univ. Paris , volume=. 2008 , publisher=

    2008

  58. [58]

    arXiv preprint arXiv:1902.10565 , year=

    Accelerating self-play learning in go , author=. arXiv preprint arXiv:1902.10565 , year=

    work page arXiv 1902

  59. [59]

    Monte Carlo Search Algorithms for Network Traffic Engineering , booktitle =

    Chen Dang and Cristina Bazgan and Tristan Cazenave and Morgan Chopin and Pierre. Monte Carlo Search Algorithms for Network Traffic Engineering , booktitle =

  60. [60]

    GPU Specs Database

    Techpowerup , HOWPUBLISHED = ". GPU Specs Database

  61. [61]

    NVIDIA , YEAR=. CUDA C

  62. [62]

    Microarchitecture of Intel AMD and VIA CPUs, Instruction Tables , NOTE =

    Agner Fog , YEAR=. Microarchitecture of Intel AMD and VIA CPUs, Instruction Tables , NOTE =

  63. [63]

    Benchmarking the Memory Hierarchy of Modern GPUs

    Mei, Xinxin and Zhao, Kaiyong and Liu, Chengjian and Chu, Xiaowen. Benchmarking the Memory Hierarchy of Modern GPUs. Network and Parallel Computing. 2014

    2014

  64. [64]

    Performance Analysis of Systems and Software (ISPASS), 2010 IEEE International Symposium on , pages=

    Demystifying GPU microarchitecture through microbenchmarking , author=. Performance Analysis of Systems and Software (ISPASS), 2010 IEEE International Symposium on , pages=. 2010 , organization=

    2010

  65. [65]

    2020 , PDF =

    Cazenave, Tristan and Lucas, Jean-Yves and Kim, Hyoseok and Triboulet, Thomas , URL =. 2020 , PDF =

    2020

  66. [66]

    Designing

    Bonnet,. Designing. Journal of Computational Biology , volume=. 2020 , publisher=

    2020

  67. [67]

    International Conference on Computer Aided Systems Theory , pages=

    Decision diagram based limited discrepancy search for a job sequencing problem , author=. International Conference on Computer Aided Systems Theory , pages=. 2019 , organization=

    2019

  68. [68]

    ICGA Journal , volume=

    Searching solitaire in real time , author=. ICGA Journal , volume=. 2007 , publisher=

    2007

  69. [69]

    , author=

    Learning Domain-Specific Control Knowledge from Random Walks. , author=. ICAPS , pages=

  70. [70]

    1996 , publisher=

    Neuro-dynamic programming , author=. 1996 , publisher=

    1996

  71. [71]

    Advances in Neural Information Processing Systems , volume=

    Solitaire: Man versus machine , author=. Advances in Neural Information Processing Systems , volume=

  72. [72]

    Journal of Heuristics , volume=

    Rollout algorithms for stochastic scheduling problems , author=. Journal of Heuristics , volume=. 1999 , publisher=

    1999

  73. [73]

    Journal of Heuristics , volume=

    Rollout algorithms for combinatorial optimization , author=. Journal of Heuristics , volume=. 1997 , publisher=

    1997

  74. [74]

    Advances in Neural Information Processing Systems , volume=

    On-line policy improvement using Monte-Carlo search , author=. Advances in Neural Information Processing Systems , volume=

  75. [75]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Combining reinforcement learning and constraint programming for combinatorial optimization , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  76. [76]

    Harvey and Matthew L

    William D. Harvey and Matthew L. Ginsberg , title =

  77. [77]

    Proceedings of the 48th Annual Southeast Regional Conference , pages=

    The snake-in-the-box problem , author=. Proceedings of the 48th Annual Southeast Regional Conference , pages=

  78. [78]

    International Conference on Principles and Practice of Constraint Programming , pages=

    Redundant modeling for the quasigroup completion problem , author=. International Conference on Principles and Practice of Constraint Programming , pages=. 2003 , organization=

    2003

  79. [79]

    Proceedings of the National Academy of Sciences , volume=

    On the falsity of Euler's conjecture about the non-existence of two orthogonal Latin squares of order 4t+ 2 , author=. Proceedings of the National Academy of Sciences , volume=. 1959 , publisher=

    1959

  80. [80]

    Comptes-Rendus de l'Association Française pour l'Avancement des Sciences , volume=

    Le problème d'Euler des 36 officiers , author=. Comptes-Rendus de l'Association Française pour l'Avancement des Sciences , volume=

Showing first 80 references.