pith. sign in

arxiv: 2606.17327 · v1 · pith:SSZNLLKInew · submitted 2026-06-15 · 🧬 q-bio.BM · cs.AR· cs.ET· cs.NE

Energy-efficient codon optimization on thermodynamic hardware

Andraz Jelincic , Ross C. Walker This is my paper

Pith reviewed 2026-06-27 02:14 UTC · model grok-4.3

classification 🧬 q-bio.BM cs.ARcs.ETcs.NE
keywords codon optimizationthermodynamic computingIsing modelPotts modelmRNA designenergy efficiencypharmaceutical computation
0
0 comments X

The pith

Codon optimization reduces to Ising or Potts model sampling executable on thermodynamic hardware with comparable quality but 10^6 lower energy use.

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

The paper reformulates the combinatorial choice of codons for an mRNA sequence as a sampling task from an Ising or Potts model. This mapping allows direct execution on a thermodynamic sampling unit that uses physical thermal fluctuations. On the SARS-CoV-2 spike protein the sampled solutions reach optimization scores similar to those from a genetic algorithm baseline. Hardware-based energy models then project that the thermodynamic unit would consume roughly a million times less energy than a conventional GPU for the same task. The work supplies the first concrete pharmaceutical application with energy numbers anchored in prototype measurements.

Core claim

mRNA codon optimization maps to sampling from an Ising or Potts model that runs on a thermodynamic sampling unit, delivering solution scores of 234-240 on the SARS-CoV-2 spike protein that match a genetic algorithm while requiring approximately 10^6 times less energy according to validated hardware models.

What carries the argument

The explicit reduction of codon selection and pairwise interaction scores to the Hamiltonian terms of an Ising or Potts model, whose thermal samples are drawn directly by the thermodynamic sampling unit.

If this is right

  • Codon optimization and similar sequence design tasks become executable on thermodynamic hardware with major energy reductions.
  • Pharmaceutical R&D workflows that rely on repeated combinatorial sampling can shift to lower-energy substrates.
  • The same mapping technique applies to any combinatorial problem already expressed as an Ising or Potts model.

Where Pith is reading between the lines

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

  • The reduction may extend to other sequence optimization problems such as protein variant design or synthetic biology circuit tuning.
  • Energy scaling behavior should be checked on sequences substantially longer than the spike protein to confirm the reported advantage persists.
  • If solution quality remains comparable across diverse proteins, thermodynamic units could replace GPU clusters for large-scale mRNA design campaigns.

Load-bearing premise

Sampling from the Ising or Potts model produces codon assignments whose optimization scores equal those obtained from established genetic algorithms.

What would settle it

Execution of the identical SARS-CoV-2 spike codon task on a physical thermodynamic sampling unit prototype with direct measurement of energy consumed and comparison of final scores to the genetic algorithm baseline.

Figures

Figures reproduced from arXiv: 2606.17327 by Andraz Jelincic, Ross C. Walker.

Figure 1
Figure 1. Figure 1: Thermodynamic computing overview. (a) A picture of a system containing two of Ex￾tropic’s prototype thermodynamic computing chips. These chips implement probabilistic computing prim￾itives and are used to confirm the correct operation of these primitives and to measure their energy consumption. (b) Schematic of a thermodynamic sampling unit (TSU). The chip contains a grid of in￾terconnected probabilistic b… view at source ↗
Figure 2
Figure 2. Figure 2: Energy comparison across methods and hardware. All methods achieve comparable optimization quality (score ∼235–243); the differen￾tiator is energy consumption, which spans approx￾imately 10 orders of magnitude. Thermodynamic chip estimates are based on validated hardware mod￾els Jelinˇciˇc et al. [2025]. Several features of this comparison are noteworthy. First, even at idealized peak efficiency (a generou… view at source ↗
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
read the original abstract

The growing energy demand for computation is becoming increasingly unsustainable. Thermodynamic computing, which harnesses physical thermal fluctuations as a computational resource rather than suppressing them, offers orders-of-magnitude energy savings for probabilistic and combinatorial tasks. Pharmaceutical R&D, heavily reliant on computational optimization and sampling, is a natural application domain. Here we present what is, to our knowledge, the first concrete pharmaceutical application mapped to thermodynamic hardware with energy estimates grounded in prototype measurements. We reduce mRNA codon optimization, a combinatorial problem routinely solved in drug development, to sampling from an Ising model, making it directly executable on a thermodynamic sampling unit (TSU). Benchmarking three approaches (Potts sampling, Ising sampling, and a genetic algorithm baseline) on the SARS-CoV-2 spike protein, we find that all achieve comparable optimization quality (scores ~234-240), but energy estimates based on validated hardware models indicate that a TSU could solve this problem using approximately 10e6 times less energy than a conventional GPU. All code is released under an open-source license.

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 / 1 minor

Summary. The paper reduces mRNA codon optimization to sampling from Ising and Potts models for direct execution on a thermodynamic sampling unit (TSU). Benchmarking on the SARS-CoV-2 spike protein shows Potts sampling, Ising sampling, and a genetic algorithm baseline achieve comparable optimization quality (scores ~234-240). Energy estimates grounded in validated hardware models from prototype measurements indicate a TSU could solve the problem using approximately 10^6 times less energy than a conventional GPU. All code is released open-source.

Significance. If the energy estimates hold, this constitutes a concrete pharmaceutical application of thermodynamic computing with potential for large efficiency gains in optimization tasks common to drug development. The open-source code release strengthens reproducibility and allows independent verification of the mapping and benchmarks.

major comments (1)
  1. [Energy estimation section] Energy estimation section: The central claim of a 10^6-fold energy reduction is derived from external validated hardware models rather than direct TSU execution or an explicit internal derivation; the manuscript should provide the specific parameters (e.g., energy per sample, number of samples required for the codon problem size, and how the Ising/Potts variable count maps to GPU baseline) used to arrive at this factor so the scaling can be verified.
minor comments (1)
  1. [Abstract] Abstract: The statement that this is 'to our knowledge, the first concrete pharmaceutical application' would be strengthened by a short note on the scope of the literature search performed.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comment on the energy estimation section. We address the request for explicit parameters below and will revise the manuscript accordingly to enhance verifiability.

read point-by-point responses

  1. Referee: [Energy estimation section] Energy estimation section: The central claim of a 10^6-fold energy reduction is derived from external validated hardware models rather than direct TSU execution or an explicit internal derivation; the manuscript should provide the specific parameters (e.g., energy per sample, number of samples required for the codon problem size, and how the Ising/Potts variable count maps to GPU baseline) used to arrive at this factor so the scaling can be verified.

    Authors: We agree that providing the specific parameters will improve transparency. The estimates rely on validated hardware models from prototype measurements (as cited in the manuscript's hardware references), not direct full-scale TSU runs, which are not yet feasible. In the revised manuscript we will add a dedicated paragraph or table in the energy estimation section explicitly listing: (1) energy per sample from prototype data, (2) number of samples required for convergence on the SARS-CoV-2 spike problem (from our benchmark runs), (3) Ising/Potts variable counts and their mapping to the codon problem size, and (4) the GPU baseline energy per sample or total used for the comparison. This will enable direct verification of the ~10^6 factor. We note that the models are externally validated rather than internally derived from first principles, which is standard for emerging hardware. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper reduces codon optimization to Ising/Potts sampling and reports benchmark scores on the SARS-CoV-2 spike protein that are numerically comparable across Potts sampling, Ising sampling, and a genetic algorithm baseline. Energy estimates are stated to rest on validated hardware models derived from prototype measurements. No internal inconsistency, unstated assumption about scaling, or mismatch between the mapped objective and the reported scores is apparent that would invalidate the central energy-savings claim. The derivation chain is self-contained against external benchmarks with no self-definitional reductions, fitted inputs called predictions, or load-bearing self-citations.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that codon optimization quality is preserved under the Ising/Potts reduction and on the accuracy of external hardware energy models; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Codon optimization can be faithfully reduced to sampling from an Ising or Potts model while retaining comparable solution quality to genetic algorithms.
    This premise enables direct execution on the TSU and underpins the quality comparison.

pith-pipeline@v0.9.1-grok · 5715 in / 1427 out tokens · 61282 ms · 2026-06-27T02:14:13.766158+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

20 extracted references · 2 canonical work pages

  1. [1]

    2024 , url =

    Electricity 2024: Analysis and Forecast to 2026 , institution =. 2024 , url =

    2024

  2. [2]

    Communications of the ACM , volume =

    Hooker, Sara , title =. Communications of the ACM , volume =. 2021 , month = nov, doi =

    2021

  3. [3]

    Jelinˇ ciˇ c, O

    Jelin. An efficient probabilistic hardware architecture for diffusion-like models , year =. doi:10.48550/arXiv.2510.23972 , url =. 2510.23972 , archiveprefix =

    work page doi:10.48550/arxiv.2510.23972

  4. [4]

    Conte, Tom and DeBenedictis, Erik and Ganesh, Natesh and Hylton, Todd and Strachan, John Paul and Williams, R. Stanley and Alemi, Alexander and Altenberg, Lee and Crooks, Gavin and Crutchfield, James and del Rio, Lidia and Deutsch, Josh and DeWeese, Michael and Douglas, Khari and Esposito, Massimiliano and Frank, Michael and Fry, Robert and Harsha, Peter ...

    work page doi:10.48550/arxiv.1911.01968 2019

  5. [5]

    IS4SI 2019 Summit , series =

    Hylton, Todd , title =. IS4SI 2019 Summit , series =. 2020 , month = jun, doi =

    2019

  6. [6]

    and Branson, Kim M

    Fox, Dillion M. and Branson, Kim M. and Walker, Ross C. , editor =. m. PLOS ONE , volume =. 2021 , month = oct, doi =

    2021

  7. [7]

    Trends in Biotechnology , volume =

    Gustafsson, Claes and Govindarajan, Sridhar and Minshull, Jeremy , title =. Trends in Biotechnology , volume =. 2004 , month = jul, doi =

    2004

  8. [8]

    Quax, Tessa E. F. and Claassens, Nico J. and S. Codon Bias as a Means to Fine-Tune Gene Expression , journal =. 2015 , month = jul, doi =

    2015

  9. [9]

    Synonymous Codons: Choose Wisely for Expression , journal =

    Brul. Synonymous Codons: Choose Wisely for Expression , journal =. 2017 , month = apr, doi =

    2017

  10. [10]

    and Gojobori, T

    Nakamura, Y. and Gojobori, T. and Ikemura, T. , title =. Nucleic Acids Research , volume =. 2000 , month = jan, doi =

    2000

  11. [11]

    In Silico Biology , volume =

    Sandhu, Kuljeet Singh and Pandey, Sunil and Maiti, Souvik and Pillai, Beena , title =. In Silico Biology , volume =. 2008 , doi =

    2008

  12. [12]

    and Thomas, Stephen J

    Polack, Fernando P. and Thomas, Stephen J. and Kitchin, Nicholas and Absalon, Judith and Gurtman, Alejandra and Lockhart, Stephen and Perez, John L. and P. Safety and Efficacy of the. New England Journal of Medicine , volume =. 2020 , month = dec, doi =

    2020

  13. [13]

    Quantum Science and Technology , volume =

    Chancellor, Nicholas , title =. Quantum Science and Technology , volume =. 2019 , month = aug, doi =

    2019

  14. [14]

    Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume =

    Berwald, Jesse and Chancellor, Nicholas and Dridi, Raouf , title =. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume =. 2022 , month = dec, doi =

    2022

  15. [15]

    IEEE Transactions on Quantum Engineering , volume =

    Chen, Jie and Stollenwerk, Tobias and Chancellor, Nicholas , title =. IEEE Transactions on Quantum Engineering , volume =. 2021 , doi =

    2021

  16. [16]

    Frontiers in Physics , volume =

    Lucas, Andrew , title =. Frontiers in Physics , volume =. 2014 , doi =

    2014

  17. [17]

    Wu, F. Y. , title =. Reviews of Modern Physics , volume =. 1982 , month = jan, doi =

    1982

  18. [18]

    and Peres, Yuval and Wilmer, Elizabeth L

    Levin, David A. and Peres, Yuval and Wilmer, Elizabeth L. , title =. 2009 , isbn =

    2009

  19. [19]

    and Gelatt, C

    Kirkpatrick, S. and Gelatt, C. D. and Vecchi, M. P. , title =. Science , volume =. 1983 , month = may, doi =

    1983

  20. [20]

    and Theogarajan, Luke , title =

    Whitehead, William and Nelson, Zachary and Camsari, Kerem Y. and Theogarajan, Luke , title =. Nature Electronics , volume =. 2023 , month = nov, doi =

    2023