Pith. sign in

REVIEW 5 major objections 4 minor 1 cited by

A New Deep-learning-Based Approach For mRNA Optimization: High Fidelity, Computation Efficiency, and Multiple Optimization Factors

T0 review · 5 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read RNop optimizes mRNA with exact protein fidelity, multi-factor losses, and throughput up to 47.32 sequences per second, with in vivo expression gains up to 4.6 times.

desk verdict A genuinely new multi-loss mRNA optimization framework, but the headline superiority claim is untested because the paper never benchmarks against existing methods. read the letter →

arxiv 2505.23862 v1 pith:6QG4IJWO submitted 2025-05-29 q-bio.QM cs.AIcs.LG

Zheng Gong , Ziyi Jiang , Weihao Gao , Deng Zhuo , Lan Ma This is my paper
classification q-bio.QMcs.AIcs.LG
keywords mRNAoptimizationcodondeeplearningsequencefidelityadaptationindextminimumfreeenergyVisionTransformer
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

This paper tries to establish that a single deep-learning model can escape the 'impossible trinity' of mRNA optimization: keeping the encoded protein exactly unchanged, running fast enough for large-scale use, and optimizing several expression-relevant sequence properties at once. The authors introduce RNop, a Transformer-based sequence-to-sequence model trained with four purpose-built losses that reward species-specific codon adaptation, tRNA availability, lower predicted minimum free energy, and exact preservation of the amino acid sequence. They report that optimized sequences improve CAI, tAI, and MFE in silico, run at up to 47.32 sequences per second, and raise protein expression in cells, including up to 4.6 times for functional proteins. A sympathetic reader would take the paper's contribution to be a practical demonstration that fidelity, speed, and multi-objective optimization are jointly achievable in one neural architecture, not a claim that any single biological index fully determines expression.

What carries the argument

The load-bearing object is the RNop architecture plus its four-loss training scheme. A fixed coding rule maps each codon to one of 64 (or 125 with ambiguity symbols) vocabulary entries, pads to a fixed length, and reshapes the feature matrix into an image-like tensor so a Vision Transformer with 16 shifted-window attention blocks can process the whole coding sequence. Each loss is a differentiable linear-algebra formulation over that coding space: GPLoss is the fidelity gate, CAILoss and tAILoss encode species-specific codon and tRNA tables, and MFELoss is built on a small 12-layer MLP that predicts minimum free energy so that gradient descent can push predicted MFE below the original sequence's value scaled by a factor of 1.1.

What would settle it

Run the trained optimizer on a held-out test set, compute the true minimum free energy of every output with the same folding software used to label training data, and plot it against the surrogate's prediction; if true MFE rises while predicted MFE falls, or the surrogate shows systematic bias on optimized sequences, the secondary-structure claim is false.

Watch

Extended reading notes

Core claim

The central claim is that RNop surpasses current mRNA optimization methods on both quantitative metrics and experimental validation. RNop encodes codons into fixed image-like matrices, processes them with a Vision Transformer, and is trained so that four losses act at once: GPLoss penalizes non-synonymous codon changes and can drive the mutation rate to zero, CAILoss pushes codon usage toward the host species' preferred codons, tAILoss favors codons matched to abundant tRNA anticodons, and MFELoss minimizes a surrogate-predicted minimum free energy. On test sets the GPLoss weight of $10^{-1}$ gives zero length errors and zero codon errors; optimized sequences show raised CAI and tAI for human, mouse, yeast, and E. coli, lower MFE, and in vivo gains in eGFP and COVID-19 spike protein expression up to 4.6 times controls. The paper concludes that deep learning can address fidelity, efficiency, and multi-objective scope simultaneously.

Load-bearing premise

The claim that optimizing minimum free energy improves expression rests on a small neural network trained to predict a sequence's minimum free energy, and if that surrogate is biased on the sequences the optimizer actually produces, the MFE gains and the expression gains attributed to them could be an artifact.

Editorial extensions

If this is right

  • At a GPLoss weight of 10^-1, RNop reports zero length errors and zero codon errors on the test set, so users can obtain exact amino-acid fidelity without post-hoc replacement.
  • Because throughput reaches 47.32 sequences per second for 3072-length inputs, whole-transcriptome or large-library optimization becomes practical on a single GPU.
  • Species-specific CAI and tAI parameters let one trained model be redirected to human, mouse, yeast, or E. coli simply by swapping loss tables.
  • The MFELoss ablation shows lower predicted MFE only when that loss is active; for yeast, optimization without it can raise MFE above the original sequence.
  • In vivo, optimized eGFP and COVID-19 spike sequences retain function and reach up to 4.6 times higher protein expression than controls.

Reading between the lines

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

  • The paper does not report any accuracy or calibration check for the MFE surrogate on the optimizer's own output distribution; until that check is done, the MFE and stability gains should be treated as surrogate-predicted, and expression gains might be driven mostly by CAI and tAI changes.
  • If the surrogate check passes, the same coding-and-loss recipe could be extended to other differentiable mRNA properties, such as predicted degradation rates or uridine depletion, without changing the architecture.
  • The image-like codon coding is generic, so it could transfer to other sequence-to-sequence genomics tasks where a small fixed vocabulary and strict length constraints matter.
  • The GPLoss weight could in principle be used deliberately to explore controlled protein variants, but the paper keeps this as a computational capability and does not test it biologically.
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

5 major / 4 minor

Summary. The manuscript introduces RNop, a Transformer-based deep-learning method for mRNA coding-sequence optimization. The method encodes mRNA sequences into image-like feature maps, processes them with a Swin-ViT architecture, and trains with four custom loss functions (GPLoss, CAILoss, tAILoss, MFELoss) intended to preserve the encoded protein, improve codon and tRNA adaptation, and lower predicted minimum free energy. The authors report in silico improvements in CAI, tAI, and MFE, a throughput of up to 47.32 sequences/s, and an in vivo eGFP fluorescence experiment. The abstract and introduction further claim that RNop surpasses current methods and increases functional-protein expression up to 4.6-fold, including for a COVID-19 spike protein.

Significance. If the central claims were adequately supported, the work would be a valuable contribution to mRNA optimization: it proposes an explicit multi-objective loss framework, addresses sequence fidelity in a deep-learning setting, and reports promising computational throughput. The idea of training a differentiable surrogate for MFE and using it inside a loss function is interesting and potentially reusable. However, the paper's main assertion of superiority over existing optimizers is not backed by any direct comparison, and several specific claims in the abstract are not substantiated in the body. As presented, the contribution is a proof-of-concept for the loss-function family and architecture rather than a demonstrated advance over the state of the art.

major comments (5)
  1. [Abstract; Sections 4 and 5] The central claim that RNop 'surpasses current methodologies' is not supported by the reported experiments. Sections 4.1-4.3 and 5.1 compare RNop-optimized sequences only against the original sequences (Tables 1-8, Figures 5-13); no results are shown for LinearDesign [17], CDSFold [16], COSMO [21], or other baselines discussed in Section 2.1. The throughput comparison in Section 6.3 and Table 9 also relies on timings quoted from other papers rather than same-machine, matched-quality runs. Without head-to-head evaluation, the superiority claim is untested, and the reader cannot assess whether RNop offers any improvement over existing methods.
  2. [Abstract; Section 1] The abstract and Section 1 state that RNop yielded 'protein expression promotion up to 4.6 times higher than the original ones' and mention experiments on both fluorescent proteins and the COVID-19 spike protein. However, Section 5.1 presents only eGFP fluorescence data, with no 4.6-fold result and no spike-protein experiments anywhere in the paper. The claimed 4.6-fold gain and the vaccine-related experiment are therefore unsupported by the presented evidence; this is a discrepancy between the stated findings and the actual results.
  3. [Section 3.5 and Section 4.3] The MFELoss uses a 12-layer, 5.4M-parameter MLP surrogate to predict MFE, but the paper reports no accuracy, calibration, or generalization checks for this surrogate, either on held-out natural sequences or on the distribution of sequences produced by the optimizer. Since the surrogate is the training signal for MFE optimization, a bias on the optimized-sequence distribution could make the surrogate's predictions unreliable, and the MFE improvements reported in Section 4.3 (computed with RNAStructure) would not be attributable to MFELoss. The authors should provide a quantitative comparison of surrogate predictions against RNAStructure values on both validation and optimized sequences.
  4. [Section 3.7] The dataset size is inconsistent between the abstract and Section 3.7. The abstract and Section 1 describe 'over 3 million sequences,' while Section 3.7 states that 2,000,000 sequences were sampled from each of three categories (Eukaryota, Bacteria, Virus), with a 90/1/9 split, which sums to 6,000,000 total sequences (5.4M training, 60k validation, 540k test). Please clarify which number is correct and ensure the abstract and main text agree.
  5. [Section 5.1] The in vivo validation is limited to a single protein (eGFP) and lacks statistical support. The text says '3 replications are conducted' and Figures 12-13 show fluorescence intensity and spectrophotometry, but no error bars, standard deviations, confidence intervals, or statistical tests are reported. Moreover, the comparison is only against the original eGFP sequence and not against sequences produced by existing optimization tools, so the claim of 'significant increase in protein expression for functional proteins compared to controls' cannot be quantitatively assessed.
minor comments (4)
  1. [Table 1] The CER row for the 10^-4 GPLoss weight appears garbled as '32.56.0'; this should be corrected to a single numeric value.
  2. [Section 6.2.1] The text states 'The results are presented in Table??' without a table number; this placeholder should be replaced with the actual table reference (likely Table 6 or 7).
  3. [Section 3.5, Eq. (1)] The sentence defining F_mfe contains a typo: 'The F_mfe means the O_i is processed by the MFE model' should be rephrased for clarity, e.g., 'F_mfe(O_i) denotes the MFE surrogate's prediction for the i-th output sequence.'
  4. [Section 3.7] 'All sequences dated to May 1th 2024' contains a typo ('1th'); it should read 'May 1, 2024.'

Circularity Check

1 steps flagged · score 6.0 of 10

In-silico validation metrics CAI and tAI are the same quantities that CAILoss and tAILoss are constructed to optimize, so Tables 2-5 partly restate the training objectives; the eGFP experiment is the only independent readout. Missing external baselines are an evidence gap, not circularity.

  1. self definitional [Sections 3.3, 3.4, 3.9, 4.2 (Tables 2-5, Figs. 5-8)]
    "Therefore, we design the CAILoss to encourage the optimization model to generate the codons with the highest CAI indexes. ... The tAILoss function can bias the optimization model toward generating codons with the highest tRNA anticodon counts. ... Therefore, we calculate the CAI of the original and optimized sequences to compare their optimality in the target species."

    CAILoss and tAILoss are constructed from the same species-specific CAI and tAI parameters that Section 3.9 later uses as evaluation metrics (Eqs. 2-3). Minimizing CAILoss is, up to optimizer convergence and competing losses, the same operation as maximizing CAI, and minimizing tAILoss is the same as maximizing tAI. Reporting original-vs-optimized CAI/tAI in Tables 2-5 and Figs. 5-8 therefore restates the training objective rather than confirming an independent prediction: the improvement is forced by the loss definition, and no external benchmark or independent biological readout is used for these specific metrics.

full rationale

The paper's central superiority claim is not circular in the sense of resting on a self-citation chain: no uniqueness theorem or author-overlapping citation carries the argument, and the eGFP fluorescence experiment is an external biological readout. The main circularity is in the in-silico validation design. CAILoss and tAILoss are defined from the same species-specific CAI/tAI tables that Section 3.9 uses as evaluation metrics; Tables 2-5 therefore demonstrate that the training objectives were optimized, not that an independent prediction was confirmed. MFELoss is a weaker case: it optimizes a learned surrogate of RNAStructure MFE and then reports MFE computed by RNAStructure, so the result is falsifiable by the surrogate's accuracy, but the paper reports no accuracy/calibration check, and the evaluation oracle is the same program that generated the training labels. The missing head-to-head comparisons with LinearDesign, CDSFold, COSMO, or RNN/LLM baselines are a substantial evidence gap for the 'surpasses current methodologies' claim, but that is a correctness/benchmarking issue, not circularity. Separately, the reported 47.32 sequences/s is a throughput measurement, not a derivation, and is not circular. Overall, one class of claimed evidence (quantitative metric improvements) reduces by construction to the loss design, while the in vivo eGFP result provides independent but limited support; hence partial circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The contributions rest on tuned hyperparameters and a learned MFE surrogate, plus biological assumptions that CAI, tAI, and MFE move protein expression. No new physical or biological entity is postulated.

free parameters (5)
  • GPLoss weight = 1e-1
    Set to 1e-1 to force zero mutation; ablation in Table 1 shows lower weights allow errors.
  • MFELoss weight multiplier = 0.01 relative to CAI/tAI
    Chosen because MFE values are O(1e2) while other losses are in (0,1]; no systematic tuning is shown.
  • MFELoss ratio lambda = 1.1
    Tuned in Table 8 as the best balance; values above 1.25 cause length and codon errors.
  • CAI/tAI loss weight multipliers = 1x for final models
    Ablation in Section 6.2.1 shows multipliers from 0.01x to 100x affect CAI, tAI, and MFE, with 100x causing unstable convergence.
  • MFE surrogate MLP parameters = 12 layers, 5.4M parameters
    Trained on RNAStructure-computed MFEs; no accuracy is reported for this surrogate.
assumptions (4)
  • domain assumption Lower MFE improves mRNA stability and half-life and therefore protein expression.
    Invoked in Sections 1 and 3.5 with citations [9,10,58]; used as justification for MFELoss.
  • domain assumption CAI and tAI are valid proxies for translation efficiency.
    Used throughout as optimization targets and evaluation metrics, following [4,6,7].
  • domain assumption The MFE surrogate generalizes to sequences produced by the optimizer.
    Section 3.5 builds MFELoss on the surrogate; no validation of surrogate accuracy on optimized sequences is provided.
  • domain assumption Sequence-to-sequence optimization on natural CDS data with these losses produces transferable optimized sequences.
    The model is trained on NCBI CDS sequences and then applied to unseen proteins like eGFP; no analysis of distribution shift is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A New Deep-learning-Based Approach For mRNA Optimization: High Fidelity, Computation Efficiency, and Multiple Optimization Factors." pith.science (2026). https://pith.science/paper/6QG4IJWO

@misc{pith2026250523862,
  author       = {Pith},
  title        = {Pith review of: A New Deep-learning-Based Approach For mRNA Optimization: High Fidelity, Computation Efficiency, and Multiple Optimization Factors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6QG4IJWO}},
  note         = {Machine review of arXiv:2505.23862}
}
read the original abstract

The mRNA optimization is critical for therapeutic and biotechnological applications, since sequence features directly govern protein expression levels and efficacy. However, current methods face significant challenges in simultaneously achieving three key objectives: (1) fidelity (preventing unintended amino acid changes), (2) computational efficiency (speed and scalability), and (3) the scope of optimization variables considered (multi-objective capability). Furthermore, existing methods often fall short of comprehensively incorporating the factors related to the mRNA lifecycle and translation process, including intrinsic mRNA sequence properties, secondary structure, translation elongation kinetics, and tRNA availability. To address these limitations, we introduce \textbf{RNop}, a novel deep learning-based method for mRNA optimization. We collect a large-scale dataset containing over 3 million sequences and design four specialized loss functions, the GPLoss, CAILoss, tAILoss, and MFELoss, which simultaneously enable explicit control over sequence fidelity while optimizing species-specific codon adaptation, tRNA availability, and desirable mRNA secondary structure features. Then, we demonstrate RNop's effectiveness through extensive in silico and in vivo experiments. RNop ensures high sequence fidelity, achieves significant computational throughput up to 47.32 sequences/s, and yields optimized mRNA sequences resulting in a significant increase in protein expression for functional proteins compared to controls. RNop surpasses current methodologies in both quantitative metrics and experimental validation, enlightening a new dawn for efficient and effective mRNA design. Code and models will be available at https://github.com/HudenJear/RPLoss.

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. Smooth $\%$MinMax: A Differentiable Relaxation for Codon Harmonization

    q-bio.QM 2026-07 conditional novelty 6.0 of 10

    Smooth %MinMax replaces discrete codon usage and hard min/max branching with probability-weighted averages and a sigmoid gate, enabling gradient-based neural codon harmonization that approximates the hard metric.

Reference graph

Works this paper leans on

62 extracted references · 46 canonical work pages · cited by 1 Pith paper

  1. [17]

    Nature621(7978), 396–403 (2023) https://doi.org/10.1038/s41586-023-06127-z

    Zhang, H., Zhang, L., Lin, A., Xu, C., Li, Z., Liu, K., Liu, B., Ma, X., Zhao, F., Jiang, H., Chen, C., Shen, H., Li, H., Mathews, D.H., Zhang, Y., Huang, L.: Algorithm for optimized mrna design improves stability and immunogenicity. Nature621(7978), 396–403 (2023) https://doi.org/10.1038/s41586-023-06127-z

    doi:10.1038/s41586-023-06127-z 2023

  2. [16]

    Bioin- formatics (Oxford, England)32(6), 828–834 (2016) https://doi.org/10.1093/ bioinformatics/btv678

    Terai, G., Kamegai, S., Asai, K.: Cdsfold: an algorithm for designing a protein-coding sequence with the most stable secondary structure. Bioin- formatics (Oxford, England)32(6), 828–834 (2016) https://doi.org/10.1093/ bioinformatics/btv678

    work page 2016

  3. [21]

    Computational and Structural Biotechnology Journal18, 1811–1818 (2020) https://doi.org/10.1016/j.csbj.2020.06.035

    Taneda, A., Asai, K.: Cosmo: A dynamic programming algorithm for multicriteria codon optimization. Computational and Structural Biotechnology Journal18, 1811–1818 (2020) https://doi.org/10.1016/j.csbj.2020.06.035

    work page doi:10.1016/j.csbj.2020.06.035 2020

  4. [1]

    BMC Bioinformatics24(1), 132 (2023) https://doi.org/10.1186/s12859-023-05246-8

    Jain, R., Jain, A., Mauro, E., LeShane, K., Densmore, D.: Icor: improving codon optimization with recurrent neural networks. BMC Bioinformatics24(1), 132 (2023) https://doi.org/10.1186/s12859-023-05246-8

    work page doi:10.1186/s12859-023-05246-8 2023

  5. [2]

    Front Bioeng Biotechnol12, 1371596 (2024)

    Paremskaia, A.I., Kogan, A.A., Murashkina, A., Naumova, D.A., Satish, A., Abramov, I.S., Feoktistova, S.G., Mityaeva, O.N., Deviatkin, A.A., Volchkov, P.Y.: Codon-optimization in gene therapy: promises, prospects and challenges. Front Bioeng Biotechnol12, 1371596 (2024)

    work page 2024

  6. [3]

    Molecular Cell61(3), 341–351 (2016) https:// doi.org/10.1016/j.molcel.2016.01.008

    Buhr, F., Jha, S., Thommen, M., Mittelstaet, J., Kutz, F., Schwalbe, H., Rodnina, M.V., Komar, A.A.: Synonymous codons direct cotranslational folding toward different protein conformations. Molecular Cell61(3), 341–351 (2016) https:// doi.org/10.1016/j.molcel.2016.01.008

    work page doi:10.1016/j.molcel.2016.01.008 2016

  7. [4]

    Nucleic Acids Research15(3), 1281–1295 (1987) https://doi.org/10.1093/nar/15.3.1281 https://academic.oup.com/nar/article-pdf/15/3/1281/3988268/15-3-1281.pdf

    Sharp, P.M., Li, W.-H.: The codon adaptation index-a measure of direc- tional synonymous codon usage bias, and its potential applications. Nucleic Acids Research15(3), 1281–1295 (1987) https://doi.org/10.1093/nar/15.3.1281 https://academic.oup.com/nar/article-pdf/15/3/1281/3988268/15-3-1281.pdf

    work page doi:10.1093/nar/15.3.1281 1987

  8. [5]

    Elife8(2019)

    Courel, M., Cl´ ement, Y., Bossevain, C., Foretek, D., Vidal Cruchez, O., Yi, Z., B´ enard, M., Benassy, M.-N., Kress, M., Vindry, C., Ernoult-Lange, M., Antoniewski, C., Morillon, A., Brest, P., Hubstenberger, A., Roest Crollius, H., Standart, N., Weil, D.: GC content shapes mRNA storage and decay in human cells. Elife8(2019)

    work page 2019

Show all 62 references
  1. [6]

    Nucleic Acids Res31(23), 6976–6985 (2003)

    Reis, M., Wernisch, L., Savva, R.: Unexpected correlations between gene expres- sion and codon usage bias from microarray data for the whole escherichia coli K-12 genome. Nucleic Acids Res31(23), 6976–6985 (2003)

    2003

  2. [7]

    Nucleic Acids Research32(17), 5036–5044 (2004) https://doi.org/10.1093/nar/gkh834 https://academic.oup.com/nar/article-pdf/32/17/5036/4156320/gkh834.pdf

    Reis, M.d., Savva, R., Wernisch, L.: Solving the riddle of codon usage preferences: a test for translational selection. Nucleic Acids Research32(17), 5036–5044 (2004) https://doi.org/10.1093/nar/gkh834 https://academic.oup.com/nar/article-pdf/32/17/5036/4156320/gkh834.pdf

    2004 doi

  3. [9]

    Mathews, D.H., Sabina, J., Zuker, M., Turner, D.H.: Expanded sequence depen- dence of thermodynamic parameters improves prediction of rna secondary structure11edited by i. tinoco. Journal of Molecular Biology288(5), 911–940 (1999) https://doi.org/10.1006/jmbi.1999.2700 28

    1999

  4. [10]

    Proceedings of the National Academy of Sciences 101(19), 7287–7292 (2004) https://doi.org/10.1073/pnas.0401799101 https://www.pnas.org/doi/pdf/10.1073/pnas.0401799101

    Mathews, D.H., Disney, M.D., Childs, J.L., Schroeder, S.J., Zuker, M., Turner, D.H.: Incorporating chemical modification constraints into a dynamic programming algorithm for prediction of rna sec- ondary structure. Proceedings of the National Academy of Sciences 101(19), 7287–...

    2004 doi

  5. [11]

    G3 Genes—Genomes—Genetics5(1), 73–80 (2015) https: //doi.org/10.1534/g3.114.015099 https://academic.oup.com/g3journal/article- pdf/5/1/73/37191150/g3journal0073.pdf

    Dana, A., Tuller, T.: Mean of the typical decoding rates: A new translation efficiency index based on the analysis of ribosome pro- filing data. G3 Genes—Genomes—Genetics5(1), 73–80 (2015) https: //doi.org/10.1534/g3.114.015099 https://academic.oup.com/g3journal/article- pdf/5...

    2015 doi

  6. [12]

    Mol Ther Nucleic Acids12, 530–542 (2018)

    Vaidyanathan, S., Azizian, K.T., Haque, A.K.M.A., Henderson, J.M., Hendel, A., Shore, S., Antony, J.S., Hogrefe, R.I., Kormann, M.S.D., Porteus, M.H., McCaf- frey, A.P.: Uridine depletion and chemical modification increase cas9 mRNA activity and reduce immunogenicity without H...

    2018

  7. [13]

    NAR Genomics and Bioinformatics6(1), 028 (2024) https: //doi.org/10.1093/nargab/lqae028 https://academic.oup.com/nargab/article- pdf/6/1/lqae028/56958324/lqae028.pdf

    Vostrosablin, N., Lim, S., Gopal, P., Brazdilova, K., Parajuli, S., Wei, X., Gromek, A., Prihoda, D., Spale, M., Muzdalo, A., Greig, J., Yeo, C., Wardyn, J., Mejzlik, P., Henry, B., Partridge, A.W., Bitton, D.A.: mrnaid, an open-source platform for therapeutic mrna design and ...

    2024 doi

  8. [14]

    Bioinformatics36(16), 4508–4509 (2020) https://doi.org/10.1093/ bioinformatics/btaa558 https://academic.oup.com/bioinformatics/article- pdf/36/16/4508/50676819/btaa558.pdf

    Zulkower, V., Rosser, S.: Dna chisel, a versatile sequence optimizer. Bioinformatics36(16), 4508–4509 (2020) https://doi.org/10.1093/ bioinformatics/btaa558 https://academic.oup.com/bioinformatics/article- pdf/36/16/4508/50676819/btaa558.pdf

    2020

  9. [15]

    Accessed: 2025-02-19 (2025)

    GenScript: GenSmart™Codon Optimization Tool. Accessed: 2025-02-19 (2025). https://www.genscript.com/tools/gensmart-codon-optimization

    2025

  10. [18]

    bioRxiv (2025) https://doi.org/10.1101/2025.02.18.638918 https://www.biorxiv.org/content/early/2025/02/21/2025.02.18.638918.full.pdf

    Brixi, G., Durrant, M.G., Ku, J., Poli, M., Brockman, G., Chang, D., Gon- zalez, G.A., King, S.H., Li, D.B., Merchant, A.T., Naghipourfar, M., Nguyen, 29 E., Ricci-Tam, C., Romero, D.W., Sun, G., Taghibakshi, A., Vorontsov, A., Yang, B., Deng, M., Gorton, L., Nguyen, N., Wang,...

    2025

  11. [19]

    Sci- entific Reports10(1), 17617 (2020) https://doi.org/10.1038/s41598-020-74091-z

    Fu, H., Liang, Y., Zhong, X., Pan, Z., Huang, L., Zhang, H., Xu, Y., Zhou, W., Liu, Z.: Codon optimization with deep learning to enhance protein expression. Sci- entific Reports10(1), 17617 (2020) https://doi.org/10.1038/s41598-020-74091-z

    2020 doi

  12. [20]

    Journal of Compu- tational Biology30(1), 70–81 (2023) https://doi.org/10.1089/cmb.2021.0458 https://doi.org/10.1089/cmb.2021.0458

    Goulet, D.R., Yan, Y., Agrawal, P., Waight, A.B., Mak, A.N.-s., Zhu, Y.: Codon optimization using a recurrent neural network. Journal of Compu- tational Biology30(1), 70–81 (2023) https://doi.org/10.1089/cmb.2021.0458 https://doi.org/10.1089/cmb.2021.0458. PMID: 35727687

    2023

  13. [22]

    In: Proceedings of the 14th ACM International Conference on Bioinformatics, Computational Biol- ogy, and Health Informatics

    Babjac, A.N., Lu, Z., Emrich, S.J.: Codonbert: Using bert for sentiment analysis to better predict genes with low expression. In: Proceedings of the 14th ACM International Conference on Bioinformatics, Computational Biol- ogy, and Health Informatics. BCB ’23. Association for C...

    2023

  14. [23]

    Biochemical and Biophysical Research Communications356(1), 136–141 (2007) https://doi.org/10.1016/j.bbrc.2007.02.127

    Park, Y.S., Seo, S.W., Hwang, S., Chu, H.S., Ahn, J.-H., Kim, T.-W., Kim, D.-M., Jung, G.Y.: Design of 5‘-untranslated region variants for tunable expression in escherichia coli. Biochemical and Biophysical Research Communications356(1), 136–141 (2007) https://doi.org/10.1016/...

    2007 doi

  15. [24]

    ACS Synthetic Biology9(5), 1051–1058 (2020) https: //doi.org/10.1021/acssynbio.9b00355

    Xiao, J., Peng, B., Su, Z., Liu, A., Hu, Y., Nomura, C.T., Chen, S., Wang, Q.: Facilitating protein expression with portable 5’-utr secondary structures in bacillus licheniformis. ACS Synthetic Biology9(5), 1051–1058 (2020) https: //doi.org/10.1021/acssynbio.9b00355

    2020 doi

  16. [25]

    ACS Synthetic Biology6(3), 555–565 (2017) https://doi.org/10.1021/ acssynbio.6b00263

    Yi, J.S., Kim, M.W., Kim, M., Jeong, Y., Kim, E.-J., Cho, B.-K., Kim, B.- G.: A novel approach for gene expression optimization through native promoter and 5’ utr combinations based on rna-seq, ribo-seq, and tss-seq of streptomyces coelicolor. ACS Synthetic Biology6(3), 555–56...

    2017

  17. [26]

    Scientific Reports9(1), 8338 (2019) https://doi.org/10.1038/s41598-019-44500-z

    Saito, Y., Kitagawa, W., Kumagai, T., Tajima, N., Nishimiya, Y., Tamano, K., 30 Yasutake, Y., Tamura, T., Kameda, T.: Developing a codon optimization method for improved expression of recombinant proteins in actinobacteria. Scientific Reports9(1), 8338 (2019) https://doi.org/1...

    2019 doi

  18. [27]

    ACS Symposium Series, vol

    Mathews, D.H., Andre, T.C., Kim, J., Turner, D.H., Zuker, M.: An Updated Recursive Algorithm for RNA Secondary Structure Prediction with Improved Thermodynamic Parameters. ACS Symposium Series, vol. 682, pp. 246–257. American Chemical Society, ??? (1997). https://doi.org/10.10...

    1997 doi

  19. [28]

    BMC Bioinformatics11(1), 129 (2010) https://doi.org/ 10.1186/1471-2105-11-129

    Reuter, J.S., Mathews, D.H.: Rnastructure: software for rna secondary structure prediction and analysis. BMC Bioinformatics11(1), 129 (2010) https://doi.org/ 10.1186/1471-2105-11-129

    2010 doi

  20. [29]

    Nature Communications13(1), 1271 (2022) https://doi.org/10

    Kierzek, E., Zhang, X., Watson, R.M., Kennedy, S.D., Szabat, M., Kierzek, R., Mathews, D.H.: Secondary structure prediction for rna sequences including n6- methyladenosine. Nature Communications13(1), 1271 (2022) https://doi.org/10. 1038/s41467-022-28817-4

    2022

  21. [30]

    Briefings in Bioinformatics24(1), 001 (2023) https://doi

    Gong, H., Wen, J., Luo, R., Feng, Y., Guo, J., Fu, H., Zhou, X.: Integrated mrna sequence optimization using deep learn- ing. Briefings in Bioinformatics24(1), 001 (2023) https://doi. org/10.1093/bib/bbad001 https://academic.oup.com/bib/article- pdf/24/1/bbad001/48782582/bbad001.pdf

    2023 doi

  22. [31]

    https://arxiv

    Devlin, J., Chang, M.-W., Lee, K., Toutanova, K.: BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (2019). https://arxiv. org/abs/1810.04805

    2019 arXiv

  23. [32]

    bioRxiv, 2023–09 (2023)

    Li, S., Moayedpour, S., Li, R., Bailey, M., Riahi, S., Kogler-Anele, L., Miladi, M., Miner, J., Zheng, D., Wang, J., et al.: Codonbert: Large language models for mrna design and optimization. bioRxiv, 2023–09 (2023)

    2023

  24. [33]

    Nature Machine Intelligence6(12), 1616–1625 (2024) https://doi.org/10.1038/ s42256-024-00946-z

    Yu, H., Yang, H., Sun, W., Yan, Z., Yang, X., Zhang, H., Ding, Y., Li, K.: An interpretable rna foundation model for exploring functional rna motifs in plants. Nature Machine Intelligence6(12), 1616–1625 (2024) https://doi.org/10.1038/ s42256-024-00946-z

    2024

  25. [34]

    Nature Methods21(12), 2287–2298 (2024) https: //doi.org/10.1038/s41592-024-02487-0

    Shen, T., Hu, Z., Sun, S., Liu, D., Wong, F., Wang, J., Chen, J., Wang, Y., Hong, L., Xiao, J., Zheng, L., Krishnamoorthi, T., King, I., Wang, S., Yin, P., Collins, J.J., Li, Y.: Accurate rna 3d structure prediction using a language model- based deep learning approach. Nature ...

    2024 doi

  26. [35]

    Bioinformatics Advances2(1), 016 (2022) https://doi.org/10.1093/ bioadv/vbac016 https://academic.oup.com/bioinformaticsadvances/article- pdf/2/1/vbac016/47081980/vbac016.pdf

    Badia-i-Mompel, P., V´ elez Santiago, J., Braunger, J., Geiss, C., Dim- itrov, D., M¨ uller-Dott, S., Taus, P., Dugourd, A., Holland, C.H., 31 Ramirez Flores, R.O., Saez-Rodriguez, J.: decoupler: ensemble of com- putational methods to infer biological activities from omics dat...

    2022

  27. [36]

    Nature Biotechnology42(4), 628–637 (2024) https://doi.org/10.1038/s41587-023-01830-8

    Wessels, H.-H., Stirn, A., M´ endez-Mancilla, A., Kim, E.J., Hart, S.K., Knowles, D.A., Sanjana, N.E.: Prediction of on-target and off-target activity of crispr– cas13d guide rnas using deep learning. Nature Biotechnology42(4), 628–637 (2024) https://doi.org/10.1038/s41587-023-01830-8

    2024 doi

  28. [37]

    Nucleic Acids Research51(20), 10934–10949 (2023) https://doi.org/10.1093/nar/gkad841 https://academic.oup.com/nar/article-pdf/51/20/10934/53175342/gkad841.pdf

    M¨ uller-Dott, S., Tsirvouli, E., Vazquez, M., Ramirez Flores, R.O., Badia- i-Mompel, P., Fallegger, R., T¨ urei, D., Lægreid, A., Saez-Rodriguez, J.: Expanding the coverage of regulons from high-confidence prior knowledge for accurate estimation of transcription factor activi...

    2023 doi

  29. [38]

    Genome Res29(8), 1363–1375 (2019)

    Garcia-Alonso, L., Holland, C.H., Ibrahim, M.M., Turei, D., Saez-Rodriguez, J.: Benchmark and integration of resources for the estimation of human transcription factor activities. Genome Res29(8), 1363–1375 (2019)

    2019

  30. [39]

    Nature Communications14(1), 7266 (2023)

    Wang, W., Feng, C., Han, R., Wang, Z., Ye, L., Du, Z., Wei, H., Zhang, F., Peng, Z., Yang, J.: trrosettarna: automated prediction of rna 3d structure with transformer network. Nature Communications14(1), 7266 (2023)

    2023

  31. [40]

    arXiv preprint arXiv:2204.00300 (2022)

    Chen, J., Hu, Z., Sun, S., Tan, Q., Wang, Y., Yu, Q., Zong, L., Hong, L., Xiao, J., Shen, T., et al.: Interpretable rna foundation model from unannotated data for highly accurate rna structure and function predictions. arXiv preprint arXiv:2204.00300 (2022)

    2022 arXiv

  32. [41]

    Nucleic acids research50(3), 14–14 (2022)

    Fu, L., Cao, Y., Wu, J., Peng, Q., Nie, Q., Xie, X.: Ufold: fast and accurate rna secondary structure prediction with deep learning. Nucleic acids research50(3), 14–14 (2022)

    2022

  33. [42]

    Nucleic acids research45(20), 11570–11581 (2017)

    Tan, Z., Fu, Y., Sharma, G., Mathews, D.H.: Turbofold ii: Rna structural align- ment and secondary structure prediction informed by multiple homologs. Nucleic acids research45(20), 11570–11581 (2017)

    2017

  34. [43]

    RNA22(12), 1808–1818 (2016)

    Sloma, M.F., Mathews, D.H.: Exact calculation of loop formation probability identifies folding motifs in rna secondary structures. RNA22(12), 1808–1818 (2016)

    2016

  35. [44]

    Nature Machine Intelligence4(12), 1174–1184 (2022) 32

    Wayment-Steele, H.K., Kladwang, W., Watkins, A.M., Kim, D.S., Tunguz, B., Reade, W., Demkin, M., Romano, J., Wellington-Oguri, R., Nicol, J.J.,et al.: Deep learning models for predicting rna degradation via dual crowdsourcing. Nature Machine Intelligence4(12), 1174–1184 (2022) 32

    2022

  36. [45]

    Kaggle (2020)

    Das, R., Wayment-Steele, H., Kim, D.S., Choe, C., Tunguz, B., Reade, W., Demkin, M.: OpenVaccine: COVID-19 mRNA Vaccine Degradation Prediction. Kaggle (2020). https://kaggle.com/competitions/stanford-covid-vaccine

    2020

  37. [46]

    coli based on deep learning

    Ding, Z., Guan, F., Xu, G., Wang, Y., Yan, Y., Zhang, W., Wu, N., Yao, B., Huang, H., Tuller, T.,et al.: Mpepe, a predictive approach to improve pro- tein expression in e. coli based on deep learning. Computational and Structural Biotechnology Journal20, 1142–1153 (2022)

    2022

  38. [47]

    Scientific Reports12(1), 12126 (2022)

    Diez, M., Medina-Mu˜ noz, S.G., Castellano, L.A., Silva Pescador, G., Wu, Q., Bazzini, A.A.: icodon customizes gene expression based on the codon composition. Scientific Reports12(1), 12126 (2022)

    2022

  39. [48]

    In: 2020 International Joint Conference on Neural Networks (IJCNN), pp

    Nabeel Asim, M., Imran Malik, M., Dengel, A., Ahmed, S.: K-mer neural embed- ding performance analysis using amino acid codons. In: 2020 International Joint Conference on Neural Networks (IJCNN), pp. 1–8 (2020). https://doi.org/10. 1109/IJCNN48605.2020.9206892

    2020

  40. [49]

    [cited 2025 Jan 16] (1988–)

    National Center for Biotechnology Information (NCBI): NCBI Genome Database. [cited 2025 Jan 16] (1988–). https://www.ncbi.nlm.nih.gov/datasets/genome/

    2025

  41. [50]

    https://www.genscript.com/ tools/codon-frequency-table

    GenScript: GenScript Codon Frequency Table. https://www.genscript.com/ tools/codon-frequency-table. Accessed: 2024/07/03

    2024

  42. [51]

    Nucleic Acids Res44(D1), 184–9 (2015)

    Chan, P.P., Lowe, T.M.: GtRNAdb 2.0: an expanded database of transfer RNA genes identified in complete and draft genomes. Nucleic Acids Res44(D1), 184–9 (2015)

    2015

  43. [52]

    Nature Methods 19(10), 1193–1207 (2022) https://doi.org/10.1038/s41592-022-01623-y

    Zhang, J., Fei, Y., Sun, L., Zhang, Q.C.: Advances and opportunities in rna struc- ture experimental determination and computational modeling. Nature Methods 19(10), 1193–1207 (2022) https://doi.org/10.1038/s41592-022-01623-y

    2022 doi

  44. [53]

    https://arxiv.org/abs/2010.11929

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (2021). https://arxiv.org/abs/2010.11929

    2021 arXiv

  45. [54]

    arXiv preprint arXiv:2103.14030 (2021)

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B.: Swin trans- former: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030 (2021)

    2021 arXiv

  46. [55]

    [cited 2025 Jan 16] (1988–)

    National Center for Biotechnology Information (NCBI): NCBI Virus Database. [cited 2025 Jan 16] (1988–). https://www.ncbi.nlm.nih.gov/labs/virus/vssi/#/

    2025

  47. [56]

    https: //arxiv.org/abs/1412.6980 33

    Kingma, D.P., Ba, J.: Adam: A Method for Stochastic Optimization (2017). https: //arxiv.org/abs/1412.6980 33

    2017 arXiv

  48. [57]

    https: //arxiv.org/abs/1711.05101

    Loshchilov, I., Hutter, F.: Decoupled Weight Decay Regularization (2019). https: //arxiv.org/abs/1711.05101

    2019 arXiv

  49. [58]

    Cell160(6), 1111–1124 (2015)

    Presnyak, V., Alhusaini, N., Chen, Y.-H., Martin, S., Morris, N., Kline, N., Olson, S., Weinberg, D., Baker, K.E., Graveley, B.R., Coller, J.: Codon optimality is a major determinant of mRNA stability. Cell160(6), 1111–1124 (2015)

    2015

  50. [59]

    Advances in Computers, vol

    Kothari, S.C., Oh, H.: Neural networks for pattern recognition. Advances in Computers, vol. 37, pp. 119–166. Elsevier (1993). https://doi.org/10. 1016/S0065-2458(08)60404-0 . https://www.sciencedirect.com/science/article/ pii/S0065245808604040

    1993

  51. [60]

    In: Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P.S., Vaughan, J.W

    Tolstikhin, I.O., Houlsby, N., Kolesnikov, A., Beyer, L., Zhai, X., Unterthiner, T., Yung, J., Steiner, A., Keysers, D., Uszkoreit, J., Lucic, M., Dosovitskiy, A.: Mlp-mixer: An all-mlp architecture for vision. In: Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P.S., Vaugha...

    2021

  52. [61]

    https://arxiv.org/abs/1901.02860

    Dai, Z., Yang, Z., Yang, Y., Carbonell, J., Le, Q.V., Salakhutdinov, R.: Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context (2019). https://arxiv.org/abs/1901.02860

    2019 arXiv

  53. [62]

    Journal of Machine Learning Research21(140), 1–67 (2020)

    Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., Liu, P.J.: Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research21(140), 1–67 (2020)

    2020

  54. [63]

    Journal of Computational Biology10(3-4), 419–432 (2003) https://doi.org/10

    Cohen, B., Skiena, S.: Natural selection and algorithmic design of mrna. Journal of Computational Biology10(3-4), 419–432 (2003) https://doi.org/10. 1089/10665270360688101 https://doi.org/10.1089/10665270360688101. PMID: 12935336 34

    2003 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.