Pith. sign in

REVIEW 5 major objections 7 minor 36 references

CAKL: Commutative algebra k-mer learning of genomics

T0 review · 5 major / 7 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read This paper introduces CAKL, a commutative-algebra encoding of k-mer positions that it claims outperforms five state-of-the-art sequence methods on eleven genomic datasets.

desk verdict The theory section is the real contribution; the headline accuracy claim depends on baselines inherited from the same group's unpublished preprint, so the margins are unverified. read the letter →

arxiv 2508.09406 v1 pith:ZMRTX77Q submitted 2025-08-13 q-bio.QM math.AC

classification q-bio.QMmath.AC MSC 55N3113F5592B10
keywords commutativealgebrak-merrepresentationalignment-freegenomicspersistentStanley–ReisnertheoryviralclassificationphylogeneticsgenomespaceVietoris–Ripscomplex
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 claims that a purely algebraic reading of where k-mers occur in a genome—tracking persistent facet ideals through a filtration instead of just counting k-mers—yields a competitive alignment-free distance for genomics. Across eleven datasets spanning variant identification, phylogenetic reconstruction, and viral classification, the proposed CAKL method is reported to beat five established sequence-analysis baselines, with the largest gains in viral classification. The stated reason is that CAKL captures positional and spatial information about k-mers that frequency-only methods discard. The authors further claim the method stays accurate as datasets grow and needs no sequence alignment, positioning it as a scalable alternative where alignment is expensive or unreliable. A sympathetic reader would take the central contribution to be a new mathematical bridge: persistent Stanley–Reisner invariants, objects from commutative algebra, are turned into a working genomic distance.

What carries the argument

Persistent Stanley–Reisner theory is the load-bearing object: for each filtration level of a simplicial complex, the Stanley–Reisner ideal is decomposed into facet prime ideals, and the persistence of those primes defines algebraic barcodes, facet counts, and persistent f- and h-vectors. The simplifying mechanism is Proposition 4.1/4.2: for Vietoris–Rips complexes built from a finite sequence of real numbers, all homology above degree zero vanishes because the complex admits a linear ordering of facets by minimal element, so the algebraic invariants reduce to the 1-skeleton graph. This reduction makes the construction computationally tractable and is what allows k-mer occurrence positions to

What would settle it

Rerun the five baseline methods on the same viral datasets, folds, and seeds with their optimal k values; if any baseline matches or exceeds CAKL's accuracy on the NCBI 2024 All set, the comparative claim fails. A second check: shuffle k-mer labels within each sequence and recompute the CAKL distance—if accuracy does not collapse, the algebraic positional encoding is not what carries the signal.

Watch

Extended reading notes

Core claim

The central claim is that k-mer sequences can be represented by persistent Stanley–Reisner invariants—algebraic features computed from facet ideals of Stanley–Reisner rings over a Vietoris–Rips filtration of k-mer occurrence positions—and that distances built from these features sort genomes accurately. For each k-mer, the paper collects its positions in a sequence as a one-dimensional point cloud; over a filtration radius, it tracks the facet ideals, f-vectors, h-vectors, and facet counts of the resulting simplicial complex. Because the point cloud is one-dimensional, the paper proves that all higher homology of these Vietoris–Rips complexes vanishes, so the persistent Betti numbers reduce

Load-bearing premise

The central outperformance claim assumes the baseline accuracy numbers, taken from a single prior study without rerunning those methods, are correct and directly comparable under identical evaluation conditions.

Editorial extensions

If this is right

  • If CAKL is correct, alignment-free methods can incorporate positional k-mer structure, not just frequencies, while remaining computationally simple enough for whole-genome datasets.
  • The one-dimensional vanishing theorem implies the full algebraic machinery can be computed from the graph skeleton, making the method scalable to megabase bacterial genomes.
  • Because the representation treats any finite-alphabet string as a word, the same construction applies to RNA, protein sequences, and other symbolic sequence data without algorithmic changes.
  • The reported stability across the four viral datasets from 2020 to 2024 suggests the method degrades gracefully as taxonomy is revised and datasets grow, unlike the baselines it is compared with.
  • Combining k-mer lengths 3, 4, and 5 in an ensemble is described as further improving accuracy at negligible cost, giving a practical recipe for users.

Reading between the lines

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

  • The one-dimensional vanishing result is specific to 1D point clouds; if CAKL is extended to 2D embeddings of genomes (such as chaos-game coordinates), the facet-invariant computation would lose this simplification and need other algebraic or topological tools.
  • A natural testable extension is direct protein-sequence classification with amino-acid alphabets, where the same facet-vector pipeline could be compared against modern protein language-model embeddings that already encode positional structure.
  • Because the comparison with five baselines relies entirely on previously published scores, an independent re-benchmark with per-method optimized k values on identical folds would be the decisive check on the order of the accuracy table.
  • The purity metric the paper introduces is a simple function of subtree concentration; one could adapt it to evaluate label coherence in other tree-building pipelines, but its sensitivity to tree shape and label imbalance should be tested before broad use.
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 / 7 minor

Summary. The paper proposes CAKL, an alignment-free genomic sequence representation in which each k-mer's occurrence positions in a sequence are treated as a one-dimensional point cloud, filtered by Vietoris–Rips complexes, and summarized by persistent Stanley–Reisner invariants (facet numbers, persistent f- and h-vectors). A weighted Euclidean distance on these feature vectors is then used with UPGMA for phylogenetic reconstruction and with 1-NN/5-NN for viral family classification. The method is evaluated on eleven datasets: one SARS-CoV-2 variant set, six phylogenetic benchmarks, and four NCBI virus collections. The central claim is that CAKL outperforms five state-of-the-art sequence-analysis methods and remains stable as dataset size grows. The paper also introduces a purity metric for phylogenetic tree monophyly and proves a vanishing theorem for higher Betti numbers of Vietoris–Rips complexes on one-dimensional sequences.

Significance. If the claims are fully supported, CAKL would be a novel and useful alignment-free framework with interpretable algebraic features, and the theoretical simplification in Propositions 4.1–4.2 is a genuine contribution that appears plausible and potentially valuable for practitioners. However, the paper's central comparative claim is not currently independently established. The baseline numbers for all five competing methods are adopted from the same research group's preprint [30] without rerunning them under the CAKL evaluation pipeline; the k-mer length for viral classification is selected after inspecting test performance; and the exact distance used in the experiments is not specified despite the composite distance being defined in Eq. (21). These issues are load-bearing for the headline conclusion, though they are in principle fixable.

major comments (5)
  1. [§3, Table 1, Fig. 3, Table S2] The abstract and §3 claim that CAKL outperforms five state-of-the-art methods by 4–7 percentage points, but Table 1, Fig. 3, and Table S2 state that the scores for NVM, FFP-JS, FFP-KL, FPS, and Markov are 'obtained from [30]', an unpublished same-group preprint. No baseline is rerun under the same evaluation conditions (same family filtering, k values, folds/seeds, metric computation, and nearest-neighbor protocol). If the conditions in [30] differ in any of these aspects, the reported margins may not reflect a fair comparison. Please rerun all baselines under identical conditions or provide a detailed equivalence argument plus the exact code and configurations used.
  2. [§2.4] The viral classification section states that 'we empirically observed that increasing the value of k ... generally leads to a decline in model performance' and then 'k = 4 selected for use in both tasks.' This appears to be model selection based on test-set accuracy on the same datasets used for the final comparison. Such tuning can overstate out-of-sample performance. Please specify a legitimate model-selection procedure—for example, a separate validation split, nested cross-validation, or a pre-specified k—and clarify why k = 4 is used for viral classification while k = 5 is used for variant and phylogenetic tasks.
  3. [§4.3] Eq. (21) defines a composite distance d = df + dh + dφ, but the final paragraph of §4.3 says the applications 'restrict our attention to a single type of feature representation—namely, the facet vector curves—and employ a fixed window length k.' The manuscript never states which distance was actually used for UPGMA and k-NN, nor how the weights ak,i in Eq. (20) were set, nor the filtration interval [r0, r1]. This makes the experiments irreproducible. Please specify the exact feature type, weight choice, filtration range, and whether Eq. (21) or a single-feature distance was used.
  4. [§4.5, Eq. (24)] The phylogenetic quality claims are quantified with a newly introduced purity metric, but Eqs. (23)–(24) depend on identifying 'all maximal subtrees whose leaves are exclusively labeled ℓ.' This requires a rooted tree and a clear convention for internal nodes, unlabeled branches, and mixed subtrees. Without an algorithmic definition of the traversal, different implementations can yield different purity values. Since the claimed ≥4 percentage point advantages in §3 rest on this metric, please define precisely how the maximal pure subtrees are obtained from a UPGMA dendrogram.
  5. [Data and Software Availability] The 'Data and Software Availability' section contains only a placeholder URL, not a working link. Given that baseline scores are inherited from a same-group preprint, an independent check of the reported results is currently impossible. Please provide a public repository with the source code, exact parameter settings, and the datasets or download scripts, and ensure the link is functional at the time of publication.
minor comments (7)
  1. [Table 1, Table S2] The column header 'FPS-JS' in Table 1 and 'FFP-JSm' in Table S2 appear to be typos for 'FFP-JS.'
  2. [§4.1, §4.2, §4.4] Typos: 'Stanely' should be 'Stanley' in §4.1; 'algeraic' should be 'algebraic' in §4.2; 'simplicifications' should be 'simplifications' in §4.4.
  3. [References [30]] Reference [30] is an arXiv preprint by the same research group. It should be marked as a preprint, and, if possible, the comparison should also include published or independently implemented baselines.
  4. [Eq. (20)] The notation in Eq. (20) is unclear: the outer sum uses K while k is also used for k-mer length, and D_k is introduced but not defined precisely. Clarify the indexing and the relationship between K and the fixed k used in experiments.
  5. [§4.5] The purity measure in Eq. (22) is the Simpson concentration index. Acknowledging this connection would help readers assess the novelty of the metric.
  6. [§2.3] The methods say all trees were built with UPGMA, but the text does not specify whether the UPGMA input is the CAKL distance from Eq. (20)/(21) or a different derived distance. Please connect the phylogenetic experiment description to the Methods section.
  7. [Abstract and §3] Phrases such as 'first-ever nonlinear algebraic framework' and 'ushers in a new era' are stronger than the evidence presented. Consider tempering these claims to match the scoped contribution.

Circularity Check

2 steps flagged · score 4.0 of 10

Outperformance claim relies on baseline scores imported from the same group's prior preprint [30], and the k hyperparameter is selected on the same evaluation datasets; the algebraic derivation itself is not circular.

  1. self citation load bearing [Table 1 caption; Fig. 3 caption; Section 3; Table S2 caption]
    "Table 1: Comparison of 1-NN classification accuracies of the six methods, where the scores of NVM, FFP-JSm FFP-KL, FPS and Markov are obtained from [30]."

    The abstract's central claim is that 'CAKL outperforms five state-of-the-art sequence analysis methods.' But all five competitor accuracies are not recomputed here; they are imported from [30], an unpublished preprint co-authored by two of the present authors (Hozumi and Wei). The reported 4–7 percentage-point margins in Section 3 are therefore differences against numbers produced by the same research group rather than independently regenerated baselines. The comparative claim is load-bearing on a self-citation and cannot be independently checked, especially since the Data and Software Availability section contains only a placeholder URL.

  2. fitted input called prediction [Section 2.4 (viral classification); Table 1; Fig. 3]
    "The model exhibits consistently strong performance for k = 3, 4, 5, with k = 4 selected for use in both tasks. For each dataset, stratified 5-fold cross-validation was conducted using 30 independent random seeds to obtain performance metrics."

    The hyperparameter k is selected as the value giving the best accuracy on the very NCBI datasets whose accuracies are later reported as predictive performance in Table 1 and Fig. 3. No separate validation split or nested cross-validation is described. Thus the reported accuracies are post-selection maxima on the evaluation data rather than out-of-sample predictions; the evaluation metric is partly fitted to the data it is then used to predict on.

full rationale

The mathematical core of CAKL is self-contained: Section 4 states and proves the needed algebraic facts (Propositions 4.1, 4.2, Theorem 4.3) rather than importing them as black boxes, and the persistent Stanley–Reisner construction is defined in the paper. No equation-level reduction of the form 'prediction equals input by construction' was found. However, the paper's headline comparative claim is not independently established: the five baseline methods are scored using numbers taken directly from [30], a preprint by two of the same authors, so the 'outperforms' claim rests on a self-citation chain for the comparison. Additionally, the choice of k=4 is made by looking at performance on the same NCBI datasets whose accuracies are then reported as predictions, which introduces a mild evaluate-on-training-data circularity. These issues warrant a score of 4: the central algebraic derivation has independent content, but the evaluation-based claims are not fully self-contained or independently verifiable.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical or ontological entities. The free parameters and underspecified choices (k, filtration range, feature type) are the main unknowns. The mathematical theorem about one-dimensional point clouds is a real result but is standard algebraic topology dressed in commutative algebra language.

free parameters (3)
  • k-mer length k = k=5 for phylogeny, k=4 for viral classification
    k was selected empirically: 'we empirically observed ... k=4 selected for use in both tasks' (Section 2.4). This choice depends on test performance.
  • Filtration interval [r0, r1]
    The filtration range for the persistent facet numbers is not specified in the paper, leaving an implementation choice.
  • Distance weights ak,i
    Equation (20) defines weights but the paper does not state which values are used, nor whether the composite distance (21) or only the facet-count distance is computed.
assumptions (3)
  • standard math Vietoris-Rips complexes of finite point sets on a line have trivial homology in degrees q >= 1 (Propositions 4.1 and 4.2).
    The proof uses Mayer-Vietoris and the linear ordering of points; it is a stated theorem the paper relies on for the simplification of features.
  • domain assumption Spatial distributions of k-mer occurrence positions capture biologically meaningful sequence relatedness.
    The entire method assumes that distances derived from k-mer position patterns reflect evolutionary or taxonomic relationships. This is stated implicitly in the introduction and Section 4.2.
  • domain assumption The known taxonomic labels (viral families, variant lineages, species) are correct ground truth for evaluating clustering and classification.
    Purity and accuracy are computed against these labels, including ICTV classifications that the paper notes are continuously revised.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CAKL: Commutative algebra k-mer learning of genomics." pith.science (2026). https://pith.science/paper/ZMRTX77Q

@misc{pith2026250809406,
  author       = {Pith},
  title        = {Pith review of: CAKL: Commutative algebra k-mer learning of genomics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZMRTX77Q}},
  note         = {Machine review of arXiv:2508.09406}
}
read the original abstract

Despite the availability of various sequence analysis models, comparative genomic analysis remains a challenge in genomics, genetics, and phylogenetics. Commutative algebra, a fundamental tool in algebraic geometry and number theory, has rarely been used in data and biological sciences. In this study, we introduce commutative algebra k-mer learning (CAKL) as the first-ever nonlinear algebraic framework for analyzing genomic sequences. CAKL bridges between commutative algebra, algebraic topology, combinatorics, and machine learning to establish a new mathematical paradigm for comparative genomic analysis. We evaluate its effectiveness on three tasks -- genetic variant identification, phylogenetic tree analysis, and viral genome classification -- typically requiring alignment-based, alignment-free, and machine-learning approaches, respectively. Across eleven datasets, CAKL outperforms five state-of-the-art sequence analysis methods, particularly in viral classification, and maintains stable predictive accuracy as dataset size increases, underscoring its scalability and robustness. This work ushers in a new era in commutative algebraic data analysis and learning.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 34 canonical work pages

  1. [30]

    Revealing the shape of genome space via k-mer topology

    Yuta Hozumi and Guo-Wei Wei. Revealing the shape of genome space via k-mer topology. arXiv preprint arXiv:2412.20202 , 2024. 34

  2. [1]

    Comparative genomics of the eukaryotes

    Gerald M Rubin, Mark D Yandell, Jennifer R Wortman, George L Gabor, Miklos, Catherine R Nelson, Iswar K Hariharan, Mark E Fortini, Peter W Li, Rolf Apweiler, et al. Comparative genomics of the eukaryotes. Science, 287(5461):2204–2215, 2000. 32

  3. [2]

    VISTA: computational tools for comparative genomics

    Kelly A Frazer, Lior Pachter, Alexander Poliakov, Edward M Rubin, and Inna Dubchak. VISTA: computational tools for comparative genomics. Nucleic acids research , 32(suppl 2):W273–W279, 2004

  4. [3]

    Phylogenetic analysis in molecular evolutionary genetics

    Masatoshi Nei. Phylogenetic analysis in molecular evolutionary genetics. Annual Review of Genetics, 30(1):371–403, 1996

  5. [4]

    Bellgard, Takeshi Itoh, Hidemi Watanabe, Tadashi Imanishi, and Takashi Gojobori

    Matthew I. Bellgard, Takeshi Itoh, Hidemi Watanabe, Tadashi Imanishi, and Takashi Gojobori. Dynamic evolution of genomes and the concept of genome space. Annals of the New York Academy of Sciences, 870(1):293–300, 1999

  6. [5]

    Gibson, Kevin Karplus, Weizhong Li, Rodrigo Lopez, Hamish McWilliam, Michael Remmert, Johannes S¨ oding, and et al

    Fabian Sievers, Andreas Wilm, David Dineen, Toby J. Gibson, Kevin Karplus, Weizhong Li, Rodrigo Lopez, Hamish McWilliam, Michael Remmert, Johannes S¨ oding, and et al. Fast, scalable generation of high-quality protein multiple sequence alignments using clustal omega. Molecular Systems Biology , 7(1):539, 2011

  7. [6]

    Standley

    Kazutaka Katoh and Daron M. Standley. Mafft multiple sequence alignment software version 7: improvements in performance and usability. Molecular Biology and Evolution , 30(4):772–780, 2013

  8. [7]

    Robert C. Edgar. Muscle: multiple sequence alignment with high accuracy and high through- put. Nucleic Acids Research, 32(5):1792–1797, 2004

Show all 36 references
  1. [8]

    Topological data analysis identifies emerging adaptive mutations in sars-cov-2

    Michael Bleher, Lukas Hahn, Maximilian Neumann, Juan Angel Patino-Galindo, Mathieu Carriere, Ulrich Bauer, Raul Rabadan, and Andreas Ott. Topological data analysis identifies emerging adaptive mutations in sars-cov-2. arXiv preprint arXiv:2106.07292 , 2021

  2. [9]

    Maranas, Peter K

    Juan ´Angel Pati˜ no-Galindo, Ioan Filip, Ratul Chowdhury, Costas D. Maranas, Peter K. Sorger, Mohammed AlQuraishi, and Raul Rabadan. Recombination and lineage-specific mutations linked to the emergence of sars-cov-2. Genome Medicine, 13(1):124, 2021

  3. [10]

    Editorial: Alignment-free methods in computational biology

    Susana Vinga. Editorial: Alignment-free methods in computational biology. Briefings in Bioinformatics, 15(3):341–342, 2014

  4. [11]

    Karlowski

    Andrzej Zielezinski, Susana Vinga, Jonas Almeida, and Wojciech M. Karlowski. Alignment-free sequence comparison: benefits, applications, and tools. Genome Biology, 18:1–17, 2017

  5. [12]

    Alignment-free genetic sequence comparisons: a review of recent approaches by word analysis

    Oliver Bonham-Carter, Joe Steele, and Dhundy Bastola. Alignment-free genetic sequence comparisons: a review of recent approaches by word analysis. Briefings in Bioinformatics , 15(6):890–905, 2014

  6. [13]

    Guillaume Bernard, Cheong Xin Chan, and Mark A. Ragan. Alignment-free microbial phy- logenomics under scenarios of sequence divergence, genome rearrangement and lateral genetic transfer. Scientific Reports, 6(1):28970, 2016

  7. [14]

    Girgis, Guillaume Bernard, Chris-Andre Leimeister, Kujin Tang, Thomas Dencker, Anna Katharina Lau, Sophie R¨ ohling, Jae Jin Choi, Michael S

    Andrzej Zielezinski, Hani Z. Girgis, Guillaume Bernard, Chris-Andre Leimeister, Kujin Tang, Thomas Dencker, Anna Katharina Lau, Sophie R¨ ohling, Jae Jin Choi, Michael S. Water- man, Matteo Comin, Sung-Hou Kim, Susana Vinga, Jonas S. Almeida, Cheong Xin Chan, Benjamin T. James...

  8. [15]

    Edwin Blaisdell

    B. Edwin Blaisdell. A measure of the similarity of sets of sequences not requiring sequence alignment. Proceedings of the National Academy of Sciences , 83(14):5155–5159, 1986

  9. [16]

    McIrvine

    Myron Tribus and Edward C. McIrvine. Energy and information. Scientific American , 225(3):179–190, 1971

  10. [17]

    Otu and Khalid Sayood

    Hasan H. Otu and Khalid Sayood. A new sequence distance measure for phylogenetic tree construction. Bioinformatics, 19(16):2122–2130, 2003

  11. [18]

    An Introduction to Kolmogorov Complexity and Its Applications , volume 3

    Ming Li and Paul Vit´ anyi. An Introduction to Kolmogorov Complexity and Its Applications , volume 3. Springer, 3rd edition, 2008

  12. [19]

    Chenglong Yu, Troy Hernandez, Hui Zheng, Shek-Chung Yau, Hsin-Hsiung Huang, Rong Lucy He, Jie Yang, and Stephen S.-T. Yau. Real time classification of viruses in 12 dimensions. PloS One, 8(5):e64328, 2013

  13. [20]

    He, and Stephen S.-T

    Mo Deng, Chenglong Yu, Qian Liang, Rong L. He, and Stephen S.-T. Yau. A novel method of characterizing genetic sequences: genome space with biological distance and applications. PloS One , 6(3):e17293, 2011

  14. [21]

    Joel Jeffrey

    H. Joel Jeffrey. Chaos game representation of gene structure. Nucleic Acids Research , 18(8):2163–2170, 1990

  15. [22]

    Milestones in graphical bioinformatics

    Milan Randi´ c, Marjana Noviˇ c, and Dejan Plavˇ si´ c. Milestones in graphical bioinformatics. International Journal of Quantum Chemistry , 113(22):2413–2446, 2013

  16. [23]

    Tung Hoang, Changchuan Yin, Hui Zheng, Chenglong Yu, Rong Lucy He, and Stephen S.- T. Yau. A new method to cluster dna sequences using fourier power spectrum. Journal of Theoretical Biology, 372:135–145, 2015

  17. [24]

    Changchuan Yin, Ying Chen, and Stephen S.-T. Yau. A measure of dna sequence similarity by fourier transform with applications on hierarchical clustering. Journal of Theoretical Biology , 359:18–28, 2014

  18. [25]

    Alignment-free method for dna sequence clustering using fuzzy integral similarity

    Ajay Kumar Saw, Garima Raj, Manashi Das, Narayan Chandra Talukdar, Binod Chandra Tripathy, and Soumyadeep Nandi. Alignment-free method for dna sequence clustering using fuzzy integral similarity. Scientific Reports, 9(1):3753, 2019

  19. [26]

    He, and Stephen S.-T

    Chenglong Yu, Qian Liang, Changchuan Yin, Rong L. He, and Stephen S.-T. Yau. A novel construction of genome space with biological geometry. DNA Research, 17(3):155–168, 2010

  20. [27]

    Hongyu Yu and Stephen S.-T. Yau. The optimal metric for viral genome space. Computational and Structural Biotechnology Journal , 23:2083–2096, 2024

  21. [28]

    Sims, Se-Ran Jun, Guohong A

    Gregory E. Sims, Se-Ran Jun, Guohong A. Wu, and Sung-Hou Kim. Alignment-free genome comparison with feature frequency profiles (ffp) and optimal resolutions. Proceedings of the National Academy of Sciences , 106(8):2677–2682, 2009

  22. [29]

    Persistent stanley–reisner theory

    Faisal Suwayyid and Guo-Wei Wei. Persistent stanley–reisner theory. Foundations of Data Science, page Doi: 10.3934/fods.2025009, 2025

  23. [31]

    Nan Sun, Shaojun Pei, Lily He, Changchuan Yin, Rong Lucy He, and Stephen S.-T. Yau. Ge- ometric construction of viral genome space and its applications. Computational and Structural Biotechnology Journal, 19:4226–4234, 2021

  24. [32]

    Statistical measures of dna sequence dissim- ilarity under markov chain models of base composition

    Tiee-Jian Wu, Ya-Ching Hsieh, and Lung-An Li. Statistical measures of dna sequence dissim- ilarity under markov chain models of base composition. Biometrics, 57(2):441–448, 2001

  25. [33]

    The biological and clinical significance of emerging sars-cov-2 variants

    Kaiming Tao, Philip L Tzou, Janin Nouhin, Ravindra K Gupta, Tulio de Oliveira, Sergei L Kosakovsky Pond, Daniela Fera, and Robert W Shafer. The biological and clinical significance of emerging sars-cov-2 variants. Nature Reviews Genetics , 22(12):757–773, 2021

  26. [34]

    Omicron ba

    Jiahui Chen and Guo-Wei Wei. Omicron ba. 2 (b. 1.1. 529.2): high potential for becoming the next dominant variant. The journal of physical chemistry letters , 13(17):3840–3849, 2022

  27. [35]

    Interactive tree of life (itol) v6: recent updates to the phylogenetic tree display and annotation tool

    Ivica Letunic and Peer Bork. Interactive tree of life (itol) v6: recent updates to the phylogenetic tree display and annotation tool. Nucleic Acids Research, page gkae268, 2024

  28. [36]

    Topological data analysis and topological deep learning beyond persistent homology-a review

    Zhe Su, Xiang Liu, Layal Bou Hamdan, Vasileios Maroulas, Jie Wu, Gunnar Carlsson, and Guo-Wei Wei. Topological data analysis and topological deep learning beyond persistent homology-a review. arXiv preprint arXiv:2507.19504 , 2025. 35

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.