Pith. sign in

REVIEW 4 major objections 5 minor 36 references

EHCube4P: Learning Epistatic Patterns Through Hypercube Graph Convolution Neural Network for Protein Fitness Function Estimation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Treating mutant sequences as a hypercube graph and denoising their fitness values with a wavelet transform before GCN training yields test R-squared of 0.66 versus 0.38 without denoising on a nine-site enzyme landscape.

desk verdict The architecture idea is reasonable, but the evaluation is invalid: test labels are wavelet-denoised with access to test data, so the reported R² does not measure prediction of unseen fitness. read the letter →

arxiv 2506.16921 v1 pith:IBIT4ZK3 submitted 2025-06-20 q-bio.QM

Muhammad Daud , Philippe Charton , Cedric Damour , Jingbo Wang , Frederic Cadet This is my paper · ORCID
classification q-bio.QM
keywords proteinfitnesspredictionepistasisgraphconvolutionalnetworkwaveletdenoisinghypercubesemi-supervisedregression
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

The paper tries to establish that protein fitness prediction improves when experimental noise is removed from a fitness signal before a graph neural network learns from it, and that the mutant space is naturally structured as a hypercube. The authors build EHCube4P, a two-layer graph convolution network on the hypercube H(9,2), where each vertex is a mutant sequence and edges connect single-mutation neighbors. They denoise the 419 measured TEAS enzyme fitness values with a one-dimensional discrete wavelet transform (Daubechies-3 basis) and then train the GCN with a smoothness regularizer to predict held-out and unmeasured mutants. The ablation study reports that removing the wavelet step drops test R-squared from 0.66 to 0.38 on the 4-EE product spectrum, which they interpret as evidence that denoising is the load-bearing component. If the claim is right, sparse and noisy fitness datasets could be modeled more reliably, easing the search for high-fitness protein variants.

What carries the argument

The central machinery is the hypercube graph H(k,2) used as the domain for a fitness signal: each vertex is a binary sequence of length k (wild-type vs mutant at each site), edges connect single-mutation neighbors, and edge weights are 1/$2^{{i⊙j}}$, reflecting interaction order. Onto this graph the paper places two GCN layers with tanh activation and normalized adjacency, followed by an MLP, and learning happens only on labeled vertices with a loss that is 70 percent mean-squared error and 30 percent graph smoothness. Before training, the experimental fitness values are treated as a one-dimensional signal and denoised by a discrete wavelet transform with universal-threshold soft shrinkage of detail coefficients, which is the step the ablation shows to be load-bearing.

What would settle it

Evaluate the trained EHCube4P model directly against the raw, non-denoised experimental fitness values on the held-out 132 mutants; if agreement is markedly worse than against the denoised values, then the model is fitting an artifact of the wavelet pipeline rather than the measured fitness. A second check is to re-run the same pipeline with several different random permutations of the 419-sequence ordering; if test R-squared swings by more than the denoising gain, the ordering assumption, not epistasis, is carrying the result.

Watch

Extended reading notes

Core claim

On its own terms, the central claim is that combining wavelet-based denoising of fitness signals with a hypercube-structured GCN produces accurate fitness estimates for unmeasured mutants, and that the improvement is driven by the denoising step. Using the TEAS/HPS library of 419 experimentally characterized mutants, the model first linearizes the fitness values by Gray code ordering, applies first-level Daubechies-3 discrete wavelet thresholding with universal soft shrinkage, and then runs two GCN layers followed by an MLP. The reported test R-squared is 0.66 for 4-epi-eremophilene production versus 0.38 without denoising, with 0.50 on two other product spectra and 0.18 on minor products. The paper also shows that performance degrades as the local variance of the fitness signal increases, concluding that smoother landscapes are more predictable and that the two-layer GCN captures beyond-pairwise epistatic interactions.

Load-bearing premise

The wavelet denoising step assumes that a sensible one-dimensional ordering of the 419 measured mutants exists and that the 93 missing sequences can be ignored, but the paper never states what ordering is used or how missing values are treated.

Editorial extensions

If this is right

  • Wavelet denoising could become a standard cheap preprocessing step for fitness landscape datasets before any graph-based or even non-graph model is trained.
  • The graph structure supports semi-supervised regression: unlabeled mutants participate in message passing, so predictions can be made for the complete 2^k space from a sparse sample.
  • The observed link between signal smoothness and test R-squared implies the model could flag which fitness landscapes are too rugged to be reliably predicted, guiding experimentalists toward smoother measurable traits.
  • Two layers of aggregation offer a concrete way to model higher-order epistatic interactions without explicitly enumerating interaction terms, which would be combinatorially expensive at large k.

Reading between the lines

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

  • Because the paper does not state the ordering of the 419 sequences used for the wavelet transform, the method's dependence on that ordering is untested; since one-dimensional wavelet denoising is not permutation invariant, I would expect a random permutation of the same fitness values to change the denoised target and therefore the reported R-squared.
  • The ablation compares the full model against raw fitness, but the training target is already denoised, so the 0.38 versus 0.66 gap measures both denoising and a changed target; a fairer test would train on raw values and evaluate on raw held-out values.
  • The framework could transfer directly to other binary combinatorial settings such as antibody libraries or regulatory DNA sequences, where a Hamming or hypercube graph and a scalar phenotypic readout are similarly available.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The manuscript proposes EHCube4P, a graph convolutional network for protein fitness prediction. The method represents the sequence space of k binary mutation sites as a hypercube H(k,2), applies a 1-D discrete wavelet transform to denoise the 419 measured fitness values, and trains a two-layer GCN followed by an MLP with a smoothness-regularized loss to predict fitness for all 512 vertices. The authors report test R² values of 0.66, 0.50, 0.50, and 0.18 for four enzyme activities of TEAS, and an ablation study on the 4-EE activity. The central claim is that wavelet preprocessing plus graph convolution improves generalization and captures higher-order epistatic interactions.

Significance. If the reported performance were obtained under a valid evaluation protocol, the framework would be a useful contribution to semi-supervised fitness prediction on sparse combinatorial sequence spaces. The paper has several strengths: the hypercube graph formulation is natural for this dataset, the use of unlabeled vertices in aggregation is well motivated, and the ablation study explicitly isolates wavelet denoising, smoothness, and the MLP. However, the current evaluation is compromised by the use of globally denoised test targets and by underspecified preprocessing, so the empirical claims are not yet established.

major comments (4)
  1. [Sec. 3.2, Eqs. (2)-(4), Fig. 4] The test-set evaluations are performed on wavelet-denoised activity values, and the denoising is a global operation over all 419 measured sequences. The detail coefficients in Eq. (2) and the universal threshold λ in Eq. (3) are computed from the full signal, so each denoised test target is a function of the entire dataset, including the training vertices and the other test vertices. Consequently, the reported test R² values (0.66 for 4-EE, 0.50 for 5-EA and PSD, 0.18 for MP) do not measure prediction of unseen experimental fitness; they measure agreement with a smoothed reconstruction that was built with access to the test measurements. To support the generalization claim, the authors should evaluate on raw experimental test values, or construct denoised targets using a procedure that does not use test data, and report both raw and denoised metrics.
  2. [Sec. 2.2, Definition 2, Fig. 3] The 1-D DWT is applied to M=419 measured fitness values, but the manuscript never states the ordering of these 419 sequences or how the 93 missing vertices of H(9,2) are handled. Definition 2 defines a Gray-code ordering on the full hypercube {0,1}^k, not on the measured subset; if only the measured vertices are ordered, consecutive entries can differ by more than one bit and the locality used to justify the transform is lost, while if the full 512-vertex ordering is used, the missing values must be imputed by an unstated rule. As a result, Eq. (2) and the denoised signals shown in Fig. 3 are not uniquely defined by the data as presented, and the preprocessing is not reproducible.
  3. [Sec. 3.2, Fig. 4, Table 1] All reported results come from a single train/test split using seed 40 (287 training vertices, 132 test vertices) with no repeated runs, confidence intervals, or alternative splits. Because the test set is small and the split is arbitrary, the differences among activities and the ablation comparisons in Table 1 may be split-specific rather than stable properties of the method. The authors should report mean and standard deviation over multiple seeds and, ideally, multiple splits, and compare against a model trained on raw, undenoiosed targets to place the R² values in context.
  4. [Sec. 2.4, Eq. (9), Sec. 3.2] The claim that the model 'effectively captures higher-order mutational interactions' is not directly tested. The two GCN layers aggregate information up to distance two in the hypercube, but no experiment isolates this effect from the contributions of wavelet denoising and the smoothness loss. A comparison against a pairwise model (e.g., a one-layer GCN or a linear epistasis model) on the same preprocessed targets is needed before the higher-order interaction claim can be supported.
minor comments (5)
  1. [Fig. 3] The caption contains the typo 'Denosing', and the x-axis label 'Mutants' does not state which ordering of the 419 sequences is used; this should be clarified.
  2. [Eq. (6)] Equation (6) has notation errors: the neighborhood is written as 'N⟩' and the normalization term is typeset as '√dj √di' without a division sign; the intended formula appears to be the normalized aggregation in Eq. (7).
  3. [Eq. (9)] In Eq. (9), the smoothness weight wi,j is not defined; the authors should state explicitly that it is the edge weight 1/2^(i⊙j) introduced in Sec. 2.3.
  4. [Sec. 3, Table 1] The schedule for λ_smooth is given only in Sec. 3 ('min(10.0, epoch/100)'), while Table 1 reports experiments with λ_smooth in different ranges; the table should state the schedule used for each row.
  5. [Data Availability Statement] The statement says the code will be released upon acceptance; providing the code and processed data with the revision would greatly improve reproducibility, especially given the underspecified preprocessing.

Circularity Check

1 steps flagged · score 6.0 of 10

Test R² is measured against denoised labels that are global functions of all 419 measurements, so the reported generalization reduces to fitting a data-dependent smoothed signal rather than predicting unseen experimental fitness.

  1. fitted input called prediction [Sec. 2.2 Eqs. (2)-(3); Sec. 3; Sec. 3.2 (Table 1, Fig. 4)]
    "The reconstructed denoised signal is passed as known labels for419 vertices, and the9d binary feature vector for all512 vertices is passed to the GCN model. ... The rest of the132 labeled vertices out of the total419 labeled vertices are kept as test vertices ... Fig. 4: ... The figure shows the correlation between the prediction of denoised experimental data ... λ =σ√2 logM (3), where σ = Median(|cd|)/0.6745."

    The test targets in Table 1 and Fig. 4 are not raw experimental fitness values; they are the wavelet-denoised reconstruction produced in Sec. 2.2. Eq. (3) sets the universal threshold using M=419 detail coefficients, so the threshold and the denoised value at every vertex, including the 132 test vertices, are global functions of all 419 measured sequences. Since the transform is computed before the 287/132 split, the denoised test label carries information from both training and test raw measurements, and the reported R² (0.66, 0.50, 0.18) measures agreement with a data-dependent smoothed signal, not prediction of unseen fitness.

full rationale

The hypercube representation, the GCN architecture, and the MLP head are self-contained and do not depend on a self-citation chain, so the score is not driven by citation practices. The central circularity is in the validation protocol: the paper first denoises all 419 measured values with a 1-D DWT whose threshold depends on the median of all detail coefficients (Eq. 3), then uses the reconstructed denoised values as the 'known labels' for both training and test vertices. Thus the test-set R² values reported in Table 1 and Fig. 4 are not evidence that the model predicts unseen experimental fitness; they are evidence that the model can fit a smoothed version of the very data that defined the test targets. This is a concrete reduction of the reported prediction to a data-derived transformation, warranting a score of 6.

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

The central claim depends on a set of domain assumptions about the wavelet transform preserving signal, the graph capturing epistasis, and the smoothness prior being appropriate. Several hand-chosen hyperparameters (lambda_smooth, hidden dimensions, edge weights, wavelet threshold rule) influence the results. No new entities are introduced.

free parameters (4)
  • lambda_smooth (smoothness regularization weight) = dynamic, ramping from 1.0 to 10.0; ablations use fixed 1.0, [0.01,1.0], [10.0,50.0]
    This hand-tuned scalar controls how strongly adjacent vertices are encouraged to have similar predicted fitness values. The paper states it is 'an arbitrary and dynamic scalar value' and shows it changes test R-squared by up to 0.08.
  • Wavelet universal threshold lambda = computed per dataset as sigma * sqrt(2 log M) with sigma = median(|detail coefficients|)/0.6745
    The threshold is derived from all measured fitness values, including test sequences, so it is a data-derived quantity that affects every denoised target.
  • GCN hidden dimensions and dropout = 512 -> 256, dropout 0.6
    Chosen by hand with no search or justification.
  • Edge weight formula 1/(2^{i dot j}) = applied to hypercube edges based on bit-wise dot product
    This ad hoc weighting scheme is described but not actually used in the GCN aggregation equations; it is a model choice with no biological justification.
assumptions (4)
  • domain assumption Gray code ordering provides a biologically meaningful linearization of the hypercube such that consecutive sequences differ by one mutation.
    Defined in Section 2.2 but never explicitly applied: the paper does not state that the DWT uses this ordering.
  • domain assumption Wavelet thresholding removes experimental noise while preserving epistatic signal.
    The paper claims this in Section 3.1, but no independent evidence is given that the removed detail coefficients are noise rather than genuine epistatic variation.
  • domain assumption Two layers of graph convolution propagate information beyond pairwise interactions and thereby capture higher-order epistasis.
    Stated in Sections 2.4 and 3.2; no direct measurement of epistatic orders is performed to validate the claim.
  • domain assumption Adjacent mutant sequences should have similar fitness values under the smoothness loss.
    This is the motivation for the L_smoothness term in Eq. 9, but it conflicts with the rugged landscapes the paper studies.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EHCube4P: Learning Epistatic Patterns Through Hypercube Graph Convolution Neural Network for Protein Fitness Function Estimation." pith.science (2026). https://pith.science/paper/IBIT4ZK3

@misc{pith2026250616921,
  author       = {Pith},
  title        = {Pith review of: EHCube4P: Learning Epistatic Patterns Through Hypercube Graph Convolution Neural Network for Protein Fitness Function Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IBIT4ZK3}},
  note         = {Machine review of arXiv:2506.16921}
}
abstract

Understanding the relationship between protein sequences and their functions is fundamental to protein engineering, but this task is hindered by the combinatorially vast sequence space and the experimental noise inherent in fitness measurements. In this study, we present a novel framework that models the sequence landscape as a hypercube $H(k,2)$ and integrates wavelet-based signal denoising with a graph convolutional neural network (GCN) to predict protein fitness across rugged fitness landscapes. Using a dataset of 419 experimentally measured mutant sequences of the Tobacco 5-Epi-Aristolochene Synthase (TEAS) enzyme, we preprocess the fitness signals using a 1-D discrete wavelet transform with a Daubechies-3 basis to suppress experimental noise while preserving local epistatic patterns. Our model comprises two GCN layers, allowing for beyond pairwise aggregation, followed by a multi-layer perceptron (MLP). We show that our approach, EHCube4P, generalizes well across different enzyme activity datasets and effectively captures higher-order mutational interactions. Performance varies with the ruggedness of the fitness landscape, with smoother signals yielding higher test set $r^2$ scores. These results demonstrate that combining wavelet preprocessing with graph-based deep learning enhances the robustness and generalization of fitness prediction, particularly for sparse and noisy biological datasets. The approach provides a scalable and interpretable framework for protein fitness estimation applicable to a broad range of combinatorial biological systems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 27 canonical work pages

  1. [1]

    Pervasive cooperative mutational effects on multiple catalytic enzyme traits emerge via long-range conformational dynamics

    Acevedo-Rocha CG, Li A, D’Amore L, Hoebenreich S, Sanchis J, Lubrano P, et al. Pervasive cooperative mutational effects on multiple catalytic enzyme traits emerge via long-range conformational dynamics. Nature Communications. 2021 Mar;12(1):1621. https://doi.org/10.1038/s41467-021-21833-w

    work page doi:10.1038/s41467-021-21833-w 2021

  2. [2]

    Machine learning to navigate fitness landscapes for protein engineer- ing

    Freschlin CR, Fahlberg SA, Romero PA. Machine learning to navigate fitness landscapes for protein engineer- ing. Current Opinion in Biotechnology. 2022;75:102713. https://doi.org/https://doi.org/10.1016/j.copbio. 2022.102713

    arXiv 2022

  3. [3]

    Protein engineering in the deep learning era

    Zhou B, Tan Y, Hu Y, Zheng L, Zhong B, Hong L. Protein engineering in the deep learning era. mLife. 2024;3(4):477–491. https://doi.org/https://doi.org/10.1002/mlf2.12157. https://onlinelibrary.wiley.com/doi/pdf/10.1002/mlf2.12157

    work page doi:10.1002/mlf2.12157 2024

  4. [4]

    In: Currin A, Swainston N, editors

    Cadet XF, Gelly JC, van Noord A, Cadet F, Acevedo-Rocha CG. In: Currin A, Swainston N, editors. Correction to: Learning Strategies in Protein Directed Evolution. New York, NY: Springer US; 2022. p. C1–C1. Available from: https://doi.org/10.1007/978-1-0716-2152-3_16

    work page doi:10.1007/978-1-0716-2152-3_16 2022

  5. [5]

    From thermodynamics to protein design: Diffusion models for biomolecule generation towards autonomous protein engineering

    Li W, Cadet XF, Medina-Ortiz D, Davari MD, Sowdhamini R, Damour C, et al.: From thermodynamics to proteindesign:Diffusion models forbiomoleculegenerationtowardsautonomousprotein engineering. Available from: https://arxiv.org/abs/2501.02680

    work page Pith review arXiv

  6. [6]

    Learning Epistasis and Residue Coevolution Patterns: Current Trends and Future Perspectives for Advancing Enzyme Engineering

    Wittmund M, Cadet F, Davari MD. Learning Epistasis and Residue Coevolution Patterns: Current Trends and Future Perspectives for Advancing Enzyme Engineering. ACS Catalysis. 2022 Nov;12(22):14243–14263. https://doi.org/10.1021/acscatal.2c01426

    work page doi:10.1021/acscatal.2c01426 2022

  7. [7]

    Exploring protein fitness landscapes by directed evolution

    Romero PA, Arnold FH. Exploring protein fitness landscapes by directed evolution. Nature Reviews Molecular Cell Biology. 2009 Dec;10(12):866–876. https://doi.org/10.1038/nrm2805

    doi:10.1038/nrm2805 2009

  8. [8]

    A review on multiple sequence alignment from the perspective of genetic algorithm

    Chowdhury B, Garai G. A review on multiple sequence alignment from the perspective of genetic algorithm. Genomics. 2017;109(5):419–431. https://doi.org/https://doi.org/10.1016/j.ygeno.2017.06.007

    work page doi:10.1016/j.ygeno.2017.06.007 2017

Show all 36 references
  1. [9]

    VISUALIZING FITNESS LANDSCAPES

    McCandlish DM. VISUALIZING FITNESS LANDSCAPES. Evolution. 2011 06;65(6):1544–

    2011

  2. [10]

    Learning the pattern of epistasis linking genotype and phenotype in a protein

    Poelwijk FJ, Socolich M, Ranganathan R. Learning the pattern of epistasis linking genotype and phenotype in a protein. Nature Communications. 2019 Sep;10(1):4213. https://doi.org/10.1038/s41467-019-12130-8

    2019 doi

  3. [11]

    A machine learning approach for reliable prediction of amino acid interactions and its application in the directed evolution of enantioselective enzymes

    Cadet F, Fontaine N, Li G, Sanchis J, Ng Fuk Chong M, Pandjaitan R, et al. A machine learning approach for reliable prediction of amino acid interactions and its application in the directed evolution of enantioselective enzymes. Scientific Reports. 2018 Nov;8(1):16757. https:/...

    2018 doi

  4. [12]

    Potts Hamiltonian models of protein co-variation, free energy landscapes, and evolutionary fitness

    Levy RM, Haldane A, Flynn WF. Potts Hamiltonian models of protein co-variation, free energy landscapes, and evolutionary fitness. Current Opinion in Structural Biology. 2017;43:55–62. Theory and simulation• Macromolecular assemblies. https://doi.org/https://doi.org/10.1016/j.s...

    2017 doi

  5. [13]

    Low-N protein engineering with data-efficient deep learning

    Biswas S, Khimulya G, Alley EC, Esvelt KM, Church GM. Low-N protein engineering with data-efficient deep learning. Nature Methods. 2021 Apr;18(4):389–396. https://doi.org/10.1038/s41592-021-01100-y

    2021 doi

  6. [14]

    Learning protein fitness models from evolutionary and assay- labeled data

    Hsu C, Nisonoff H, Fannjiang C, Listgarten J. Learning protein fitness models from evolutionary and assay- labeled data. Nature Biotechnology. 2022 Jul;40(7):1114–1122. https://doi.org/10.1038/s41587-021-01146-5

    2022 doi

  7. [15]

    Higher-orderepistasisandphenotypic prediction

    ZhouJ,WongMS,ChenWC,KrainerAR,KinneyJB,McCandlishDM. Higher-orderepistasisandphenotypic prediction. Proceedings of the National Academy of Sciences. 2022;119(39):e2204233119. https://doi.org/10. 1073/pnas.2204233119. https://www.pnas.org/doi/pdf/10.1073/pnas.2204233119

    2022 doi

  8. [16]

    A Gentle Introduction to Graph Neural Networks

    Sanchez-Lengeling B, Reif E, Pearce A, Wiltschko AB. A Gentle Introduction to Graph Neural Networks. Distill. 2021;Https://distill.pub/2021/gnn-intro. https://doi.org/10.23915/distill.00033. 10

    2021 doi

  9. [17]

    Graph Neural Networks and Their Current Applications in Bioinformatics

    Zhang XM, Liang L, Liu L, Tang MJ. Graph Neural Networks and Their Current Applications in Bioinformatics. Frontiers in Genetics. 2021;Volume 12 - 2021. https://doi.org/10.3389/fgene.2021.690049

    2021

  10. [18]

    ProS-GNN: Predicting effects of mutations on protein stability using graph neural networks

    Wang S, Tang H, Shan P, Wu Z, Zuo L. ProS-GNN: Predicting effects of mutations on protein stability using graph neural networks. Computational Biology and Chemistry. 2023;107:107952. https://doi.org/https: //doi.org/10.1016/j.compbiolchem.2023.107952

    2023

  11. [19]

    Quantitative exploration of the catalytic landscape separating divergent plant sesquiterpene synthases

    O’Maille PE, Malone A, Dellas N, Andes Hess B, Smentek L, Sheehan I, et al. Quantitative exploration of the catalytic landscape separating divergent plant sesquiterpene synthases. Nature Chemical Biology. 2008 Oct;4(10):617–623. https://doi.org/10.1038/nchembio.113

    2008 doi

  12. [20]

    Biosynthetic potential of sesquiterpene synthases: product profiles of Egyptian Henbane premnaspirodiene synthase and related mutants

    Koo HJ, Vickery CR, Xu Y, Louie GV, O’Maille PE, Bowman M, et al. Biosynthetic potential of sesquiterpene synthases: product profiles of Egyptian Henbane premnaspirodiene synthase and related mutants. The Journal of Antibiotics. 2016 Jul;69(7):524–533. https://doi.org/10.1038/...

    2016 doi

  13. [21]

    In: Sikosek T, editor

    Poelwijk FJ. In: Sikosek T, editor. Context-Dependent Mutation Effects in Proteins. New York, NY: Springer New York; 2019. p. 123–134. Available from: https://doi.org/10.1007/978-1-4939-8736-8_7

    2019 doi

  14. [22]

    Wavelets: Mathematical Theory

    Schneider K, Farge M. Wavelets: Mathematical Theory. In: Françoise JP, Naber GL, Tsun TS, editors. Encyclopedia of Mathematical Physics. Oxford: Academic Press; 2006. p. 426–438. Available from: https: //www.sciencedirect.com/science/article/pii/B012512666200153X

    2006

  15. [23]

    A wavelet tour of signal processing

    Mallat S. A wavelet tour of signal processing. Elsevier; 1999

    1999

  16. [24]

    A Review of Wavelet Analysis and Its Applications: Challenges and Opportunities

    Guo T, Zhang T, Lim E, López-Benítez M, Ma F, Yu L. A Review of Wavelet Analysis and Its Applications: Challenges and Opportunities. IEEE Access. 2022;10:58869–58903. https://doi.org/10.1109/ACCESS.2022. 3179517

    2022 doi

  17. [25]

    4.2 - Multiscale Image Decompositions and Wavelets

    Moulin P. 4.2 - Multiscale Image Decompositions and Wavelets. In: BOVIK A, editor. Handbook of Image and Video Processing (Second Edition). second edition ed. Communications, Networking and Multimedia. Burlington:AcademicPress;2005.p.347–359. Availablefrom:https://www.scienced...

    2005

  18. [26]

    A New Wavelet Threshold Determination Method Considering Interscale Correlation in Signal Denoising

    He C, Xing J, Li J, Yang Q, Wang R. A New Wavelet Threshold Determination Method Considering Interscale Correlation in Signal Denoising. Mathematical Problems in Engineering. 2015;2015(1):280251. https://doi. org/https://doi.org/10.1155/2015/280251. https://onlinelibrary.wiley...

    2015 doi

  19. [27]

    Ideal spatial adaptation by wavelet shrinkage

    Donoho DL, Johnstone IM. Ideal spatial adaptation by wavelet shrinkage. biometrika. 1994;81(3):425–455

    1994

  20. [28]

    Mechanisms of DNA damage, repair, and mutagenesis

    Chatterjee N, Walker GC. Mechanisms of DNA damage, repair, and mutagenesis. Environ Mol Mutagen. 2017 May;58(5):235–263

    2017

  21. [29]

    Mutation, repair and recombination

    Brown TA. Mutation, repair and recombination. New York, NY: Wiley-Liss; 2002

    2002

  22. [30]

    The Context-Dependence of Mutations: A Linkage of Formalisms

    Poelwijk FJ, Krishna V, Ranganathan R. The Context-Dependence of Mutations: A Linkage of Formalisms. PLOS Computational Biology. 2016 06;12(6):1–19. https://doi.org/10.1371/journal.pcbi.1004771

    2016 doi

  23. [31]

    A new model for learning in graph domains

    Gori M, Monfardini G, Scarselli F. A new model for learning in graph domains. In: Proceedings. 2005 IEEE international joint conference on neural networks, 2005.. vol. 2. IEEE; 2005. p. 729–734

    2005

  24. [32]

    Semi-Supervised Classification with Graph Convolutional Networks

    Kipf TN, Welling M. Semi-Supervised Classification with Graph Convolutional Networks. In: International Conference on Learning Representations; 2017. Available from: https://openreview.net/forum?id=SJU4ayYgl

    2017

  25. [33]

    A Flexible Generative Framework for Graph-based Semi-supervised Learning

    Ma J, Tang W, Zhu J, Mei Q. A Flexible Generative Framework for Graph-based Semi-supervised Learning. In: Wallach H, Larochelle H, Beygelzimer A, d'Alché-Buc F, Fox E, Garnett R, editors. Advances in Neural Information Processing Systems. vol. 32. Curran Associates, Inc.; 2019...

    2019

  26. [34]

    The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains

    Shuman DI, Narang SK, Frossard P, Ortega A, Vandergheynst P. The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains. IEEE Signal Processing Magazine. 2013;30(3):83–98. https://doi.org/10.1109/MSP.2012...

    2013

  27. [35]

    Minimum epistasis interpolation for sequence-function relationships

    Zhou J, McCandlish DM. Minimum epistasis interpolation for sequence-function relationships. Nature Communications. 2020 Apr;11(1):1782. https://doi.org/10.1038/s41467-020-15512-5. 12

    2020 doi

  28. [1558]

    https://academic.oup.com/evolut/article- pdf/65/6/1544/47949872/evolut1544.pdf

    https://doi.org/10.1111/j.1558-5646.2011.01236.x. https://academic.oup.com/evolut/article- pdf/65/6/1544/47949872/evolut1544.pdf

    2011

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.