Pith. sign in

REVIEW 4 major objections 5 minor 56 references

Homologous nodes in annotated complex networks

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

Pith's one-line read Nodes with similar neighbourhood annotations share functional roles even when unconnected.

desk verdict A simple and reproducible method for finding functionally similar nodes in annotated networks, let down by a statistically over-claimed GO enrichment analysis. read the letter →

arxiv 2505.23668 v1 pith:U2LYP4KC submitted 2025-05-29 cs.SI

classification cs.SI
keywords annotatednetworksnodeannotationsneighbourhoodconnectivityprofileshierarchicalclusteringfunctionalrolesfoodwebgeneregulatorynetworkrecipe
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 proposes that a node's role in a network can be read from the annotations attached to its neighbours: for each node, count the labels in its neighbourhood, normalise those counts, and cluster the resulting profiles. Two nodes are called homologues when their neighbourhood annotation distributions match, and they need not share an edge or belong to a dense community. Applied to three very different networks, the clusters line up with real functional categories: trophic level and predator-prey roles in a food web, cuisine-specific versus universal ingredients and geographically coherent cuisines, and Gene Ontology functions of transcription factors in a plant regulatory network. If this works, it gives network science a way to detect functionally meaningful groupings in systems where modular community structure is weak or absent, and where annotations rather than edges carry the signal.

What carries the argument

The central object is the connectivity vector $v_{ik} = \sum_j a_{ij} b_{jk}$, equivalently the matrix product $A B$, with directed variants $A^T B$ and $A B$, normalised to a probability distribution over annotation types. This vectorisation turns each node's local role into a point in annotation space, and the same matrix multiplication works for unipartite, directed, bipartite, and tripartite networks with single or multiple labels per node. All subsequent steps, such as hierarchical clustering with Euclidean distance and Ward linkage, silhouette scores, label-permutation null models, and Gene Ontology enrichment, are ways of reading and testing these vectors.

What would settle it

Sweep the tree-cut height across a range such as 0.1 to 0.5 times total height for each of the three networks and recompute the silhouette scores and GO enrichment. If nearby cuts dissolve the fish and apex-predator clusters, the Japanese and Thai ingredient clusters, or the enriched transcription-factor clusters into non-significance, the reported functional roles depend on a fragile choice.

Watch

Extended reading notes

Core claim

The paper claims that clustering nodes by the distributions of annotations in their local neighbourhoods identifies common functional roles and properties, even when the clustered nodes are not connected to one another. For directed networks the method computes separate in- and out-vectors, so a predator's profile is built from its prey's annotations and vice versa; ingredients are embedded in cuisine space and cuisines in ingredient space; transcription factors are embedded in the Gene Ontology space of their upstream regulators and downstream targets. Hierarchically clustering these vectors places organisms at similar trophic levels together, including an apex-predator group of diverse taxa, separates cuisine-specific from universal ingredients, reproduces historical and geographic cuisine groupings, and assigns 559 of 580 transcription factors to clusters enriched in specific GO terms. The paper argues that such groupings, which it calls homologues, reveal functional roles visible in local connectivity context but not captured by community detection or by simple assortative or disassortative structure.

Load-bearing premise

The reported groupings all come from slicing each clustering tree at a fixed fraction of its total height (0.2 for the food web, 0.3 for ingredients, 0.4 for cuisines), and the paper does not test whether other slice positions preserve the same groupings.

Editorial extensions

If this is right

  • Across domains, the same pipeline yields interpretable classes: ecological roles, culinary groupings, and transcription-factor functions are all read off neighbourhood annotation profiles.
  • Incomplete metadata could be supplemented, because a gene's position in the clustering suggests functions even when its own annotation is missing, a use the paper explicitly notes.
  • Because homologues are not required to be connected, the approach applies where communities are sparse or disassortative, a regime the paper argues existing modularity-based methods handle poorly.
  • Statistical significance in each case is established against label-permuted nulls or false-discovery-corrected enrichment, so the reported groupings are claimed to be unlikely under random relabelling.

Reading between the lines

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

  • A use the authors leave implicit is label imputation: an unannotated node could be placed into the cluster of its nearest neighbourhood profile and assigned the dominant annotations of that cluster.
  • The paper's two null models, shuffling annotations while keeping structure and rewiring edges while keeping labels, separate two distinct questions, and this separation could serve as a general diagnostic for whether any annotation scheme is coupled to topology.
  • The method is demonstrated on static networks only; extending the vectors to temporal or multilayer neighbourhoods is a natural next step, but the paper offers no evidence about that regime.
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 paper introduces a method for clustering nodes in annotated networks based on the distributions of node annotations in each node's neighbourhood. The authors define 'connectivity vectors' by multiplying adjacency and annotation matrices, normalise them, and apply hierarchical clustering. They name the resulting groups 'homologous nodes' and demonstrate the approach on three networks: a Florida food web, a tripartite recipe network, and an Arabidopsis thaliana gene regulatory network with Gene Ontology annotations. The central claim is that these groupings identify common functional roles even when the grouped nodes are not directly connected. The manuscript also reports silhouette-score-based permutation tests and a GO enrichment analysis for the gene regulatory network, along with released code and data.

Significance. If the central claim is robust, the method would be a useful, interpretable addition to the toolbox for annotated-network analysis: it is simple, fast, and directly connects topology with metadata. The authors are commendably explicit about providing code and data, and the vectorisation itself is parameter-free. However, the current evidence is not fully convincing. The strongest quantitative result—the Arabidopsis GO enrichment analysis—is weakened by a selection procedure that scans all dendrogram branch points without correcting for that search. The silhouette scores are close to zero, the permutation p-values are resolution-limited, and the dendrogram cut heights are chosen without sensitivity analysis. These issues do not invalidate the conceptual approach, but they do mean that the paper's central empirical claims require additional support before publication.

major comments (4)
  1. [II.C and Methods V.D] The GO enrichment evidence is compromised by selection over dendrogram branch points. The text states that the authors 'ranked all possible clusters produced by the dendrogram, in other words all possible branch points of the tree, first by the number of enriched GO terms ... and second by the size of the cluster', then discarded overlapping clusters. The Benjamini-Hochberg correction in Methods V.D is applied to the GO terms within each candidate cluster only; it does not control for the fact that roughly 580 candidate clusters were searched. Under a null hypothesis where TF function is independent of connectivity profiles, some branch points will appear enriched by chance, and the selection rule will necessarily retain the most enriched ones. Therefore the reported 96% coverage (559/580 TFs) is not, by itself, evidence of true functional signal. I recommend a global correction across all hypotheses tested at all branch points, or a permutation null that randomly shuffles TF–GO associations and repeats the full branch-scanning procedure, reporting the distribution of coverage and enrichment counts under that null.
  2. [II.A, II.B.1, II.B.2] The cluster memberships, silhouette scores, and enrichment results all depend on dendrogram cut heights that are chosen by inspection: 0.2x total height for the food web, 0.3x for recipe ingredients, and 0.4x for cuisines. No sensitivity analysis or data-driven criterion is provided. Because the paper's qualitative conclusions are based on the specific clusters obtained at these cuts, the authors should show that the main groupings and statistics are stable across a range of cut heights, or justify the chosen heights independently of the results they produce.
  3. [II.A, II.B.1, II.B.2, II.B.3; Figures 1D, 2D, 3B] The reported statistical support is fragile. The silhouette scores are near zero (s=0.1, 0.12, 0.13, and -0.20), and the permutation-based p-values are all bounded by the 1000-permutation resolution as p<0.001. Values around zero indicate essentially overlapping clusters, so a highly significant p-value against a null does not establish that the grouping is functionally meaningful. The paper also runs multiple null tests without any multiple-testing correction. I request effect sizes with confidence intervals, a finer permutation resolution (or an analytic null), and a statement of whether the conclusions survive correction for the multiple comparisons performed.
  4. [II.C and Discussion] The connection between the clustering result and the functional interpretation is partially circular in the gene regulatory network: the clusters are built from GO-term profiles of neighbours, and the same GO annotation source is used to validate the clusters via enrichment. The authors should clarify whether the GO terms that are enriched in a cluster overlap with the GO terms used to construct the vectors, and, if so, discuss what new information the enrichment analysis adds beyond the construction itself. A cleaner validation would use held-out annotations or an independent functional database.
minor comments (5)
  1. [II.A] There is a typo in 'To quantify the the extent' that should be corrected.
  2. [Appendix B / Figure A4] The text says the silhouette score is negative (s=-0.13) and 'statistically significant (p<0.001)', but it also says all random shufflings showed 'greater magnitude of disagreement'; please clarify whether the observed value is more extreme in the direction of disagreement and report what the p-value actually tests.
  3. [Methods V.E] The probabilistic graph traversal has a typo ('probability of of') and the corrective factor c is set to 1 without a sensitivity check; since c is listed as a free parameter, a brief robustness statement would be helpful.
  4. [Methods V.D and II.C] The filter that ignores GO terms appearing only once in the network is described but not analysed for its effect on results; please state how many terms are removed and whether the conclusions change with a different filter threshold.
  5. [General] Some references are formatted inconsistently (e.g., the eLife entries), and the Supplementary Information is referenced but not fully described in the main text; listing the supplementary files and their contents would improve reproducibility.

Circularity Check

1 steps flagged · score 6.0 of 10

The Arabidopsis GO-enrichment validation is circular by construction: the reported clusters are selected by ranking branch points on the number of enriched GO terms, so their enrichment is the selection criterion, not an independent confirmation of functional homology.

  1. fitted input called prediction [Section II.C and Methods V.D (Gene Ontology Enrichment Analysis).]
    "We ranked all possible clusters produced by the dendrogram, in other words all possible branch points of the tree, first by the number of enriched GO terms (priority given to larger number of enriched terms), and second by the size of the cluster (priority to larger). Then, we discarded any clusters that shared subsets of TFs with a higher-ranked cluster. The final result of this procedure is a non-overlapping set of clusters with enriched GO terms."

    The set of 'functionally enriched clusters' is the output of the extraction rule that ranks branch points by number of Benjamini-Hochberg-significant GO terms and discards overlap. Reporting these clusters as enriched is therefore restating the selection criterion, not an independent prediction from connectivity profiles. The BH correction in Methods V.D is applied to GO terms within each candidate cluster only, not across the roughly 579 branch points searched, so the search is unaccounted for. Under a null of no association between neighbourhood-annotation vectors and TF function, the same ranking-and-discarding procedure would still yield branch points with nominally enriched GO terms.

full rationale

The core vectorisation (Eqs. 1-4) is parameter-free: connectivity vectors are matrix products of the adjacency and annotation matrices, and hierarchical clustering is a standard algorithm applied to those vectors. The food-web and recipe validations use external labels (coarse organism categories, topological layers, geographical cuisine regions, manually labelled ingredient types) and compare observed silhouette scores against 1000 label-permuted null realisations; those tests are not circular. The arbitrary dendrogram cut heights (0.2x, 0.3x, 0.4x) are a robustness concern but do not by themselves make the derivation circular, because the statistical silhouette tests are computed on the vector embeddings against external groupings, not on the chosen cut clusters. Self-citations are limited to data sources and prior biological motivation (e.g., refs. 27, 31-32, 43); these are independent empirical inputs, not internal justifications. The one genuine circular step is the Arabidopsis GO analysis: clusters are selected by maximizing the number of enriched GO terms, and the same enrichment is then reported as evidence that the clusters are functionally coherent. Because the false-discovery correction does not span the branch-point search, the reported enrichment and 96% TF coverage are forced by the extraction rule. This makes the GO-based demonstration partially circular, while the method itself and the other two applications remain non-circular, yielding an overall score of 6.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The main free parameters are the dendrogram cut heights and filtering thresholds; no fitted model parameters, but the choices affect results. The axioms are standard mathematical tools plus domain assumptions about the meaning of annotations and the validity of null models.

free parameters (6)
  • Dendrogram cut height (food web) = 0.2 × total height
    Chosen to produce interpretable clusters; not theoretically justified.
  • Dendrogram cut height (recipe ingredients) = 0.3 × total height
    Chosen ad hoc to separate cuisine-specific from universal ingredients.
  • Dendrogram cut height (recipe cuisines) = 0.4 × total height
    Chosen ad hoc to show six cuisine clusters.
  • GO term frequency filter = Ignore GO terms appearing only once in the network
    Filtering choice; could affect downstream vectors and clusters.
  • Topological traversal corrective factor c = 1
    Set to 1 without sensitivity analysis.
  • Benjamini-Hochberg false discovery rate = α = 0.1
    Significance threshold chosen for GO enrichment.
assumptions (5)
  • domain assumption The annotation matrix B accurately represents node labels for each network.
    The connectivity vectors are computed from B; missing or erroneous annotations would propagate directly into the vectors and clusters.
  • domain assumption The distribution of annotations among a node's neighbours encodes its functional role.
    This is the premise that makes 'homologues' meaningful; the paper tests it indirectly via silhouette scores against predefined labels.
  • standard math Euclidean distance and Ward's method are suitable for clustering normalised probability vectors.
    Standard and widely used choices, but are not specifically justified for probability-simplex data.
  • domain assumption The permutation null models preserve the network structure and annotation frequencies needed for valid p-values.
    The nulls shuffle annotations or recipe memberships but assume exchangeability of these components; if the real dependencies are more complex, the p-values may be anticonservative.
  • standard math The hypergeometric test and Benjamini-Hochberg correction are appropriate for GO enrichment analysis.
    Standard methods used correctly; however, the cluster selection procedure itself is not included in the null model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Homologous nodes in annotated complex networks." pith.science (2026). https://pith.science/paper/U2LYP4KC

@misc{pith2026250523668,
  author       = {Pith},
  title        = {Pith review of: Homologous nodes in annotated complex networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U2LYP4KC}},
  note         = {Machine review of arXiv:2505.23668}
}
read the original abstract

Many real-world networks have associated metadata that assigns categorical labels to nodes. Analysis of these annotations can complement the topological analysis of complex networks. Annotated networks have typically been used to evaluate community detection approaches. Here, we introduce an approach that combines the quantitative analysis of annotations and network structure, which groups nodes according to similar distributions of node annotations in their neighbourhoods. Importantly the nodes that are grouped together, which we call homologues may not be connected to each other at all. By applying our approach to three very different real-world networks we show that these groupings identify common functional roles and properties of nodes in the network.

Figures

Figures reproduced from arXiv: 2505.23668 by the authors.

Figure 1
Figure 1. FIG. 1. Food web clustering and visualisation. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Clustering of the gene regulatory network using vectors derived from Gene Ontology (GO) labels. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages

  1. [1]

    Clustering reveals sets of ingredients that are characteristic for particular cuisines Next, we apply our approach to the recipe network: a tripartite network of recipes, ingredients, and cuisine annotations. We first project this multipartite network by counting the number of recipes in each cuisine asso- ciated with each ingredient, and create a cuisine...

  2. [2]

    homologous

    Clustering reveals hierarchical similarity of cuisine types In a parallel process to the construction of cuisine- space vectors of ingredients, we now assemble the ingredient-space vectors of cuisines in the following way: Using the bipartite graph matrix elementsa ij andb jk as before, the number of times an ingredientkis used in a given cuisineiisν ik =...

  3. [3]

    M. E. J. Newman,Networks, 2nd ed. (Oxford University Press, Oxford, 2018)

  4. [4]

    D. S. Bassett and O. Sporns, Nature Neuroscience20, 353 (2017), publisher: Nature Publishing Group

  5. [5]

    A. Lin, R. Yang, S. Dorkenwald, A. Matsliah, A. R. Ster- ling, P. Schlegel, S.-c. Yu, C. E. McKellar, M. Costa, K. Eichler, A. S. Bates, N. Eckstein, J. Funke, G. S. X. E. Jefferis, and M. Murthy, Nature634, 153 (2024), publisher: Nature Publishing Group

  6. [6]

    Hobert, L

    O. Hobert, L. Glenwinkel, and J. White, Current Biology 26, R1197 (2016)

  7. [7]

    Fortunato and M

    S. Fortunato and M. E. J. Newman, Nature Physics18, 848 (2022), publisher: Nature Publishing Group

  8. [8]

    M. E. J. Newman and A. Clauset, Nature Communica- tions7, 11863 (2016), number: 1 Publisher: Nature Pub- lishing Group

Show all 56 references
  1. [9]

    Newman and T

    M. Newman and T. P. Peixoto, Physical Review Letters 115, 088701 (2015), publisher: American Physical Soci- ety

  2. [10]

    M. T. Schaub, J. Li, and L. Peel, Physical Review E107, 054305 (2023), publisher: American Physical Society

  3. [11]

    Guimer` a and L

    R. Guimer` a and L. A. N. Amaral, Journal of Statisti- cal Mechanics: Theory and Experiment2005, P02001 (2005)

  4. [12]

    Guimer` a and L

    R. Guimer` a and L. A. Nunes Amaral, Nature433, 895 (2005), number: 7028 Publisher: Nature Publishing Group

  5. [13]

    Guimer` a, M

    R. Guimer` a, M. Sales-Pardo, and L. A. N. Amaral, Na- ture Physics3, 63 (2007), publisher: Nature Publishing Group

  6. [14]

    Pedersen, A

    M. Pedersen, A. Omidvarnia, J. M. Shine, G. D. Jackson, and A. Zalesky, Network Neuroscience4, 416 (2020)

  7. [15]

    Sharma, J

    A. Sharma, J. Menche, C. C. Huang, T. Ort, X. Zhou, M. Kitsak, N. Sahni, D. Thibault, L. Voung, F. Guo, S. D. Ghiassian, N. Gulbahce, F. Baribaud, J. Tocker, R. Dobrin, E. Barnathan, H. Liu, R. A. Panettieri, Jr, K. G. Tantisira, W. Qiu, B. A. Raby, E. K. Silverman, M. Vidal, ...

  8. [16]

    S. D. Ghiassian, J. Menche, and A.-L. Barab´ asi, PLOS Computational Biology11, e1004120 (2015), publisher: Public Library of Science

  9. [17]

    M. E. J. Newman, Physical Review Letters89, 208701 (2002)

  10. [18]

    Noldus and P

    R. Noldus and P. Van Mieghem, Journal of Complex Net- works3, 507 (2015)

  11. [19]

    Homophily Within and Across Groups,

    A. K. Rizi, R. Michielan, C. Stegehuis, and M. Kivel¨ a, “Homophily Within and Across Groups,” (2024), version Number: 1

  12. [20]

    M. E. J. Newman, Physical Review E67, 026126 (2003), publisher: American Physical Society

  13. [21]

    Kosch¨ utzki and F

    D. Kosch¨ utzki and F. Schreiber, Gene Regulation and Systems Biology2, GRSB.S702 (2008), publisher: SAGE Publications Ltd STM

  14. [22]

    K. Das, S. Samanta, and M. Pal, Social Network Anal- ysis and Mining8, 13 (2018)

  15. [23]

    J. Sun, J. Kunegis, and S. Staab, in2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW)(2016) pp. 128–135, iSSN: 2375-9259

  16. [24]

    Bhagat, G

    S. Bhagat, G. Cormode, and S. Muthukrishnan, inSo- cial Network Data Analytics, edited by C. C. Aggarwal (Springer US, Boston, MA, 2011) pp. 115–148

  17. [25]

    M. E. J. Newman, Nature Physics14, 542 (2018), pub- lisher: Nature Publishing Group

  18. [26]

    Henderson, B

    K. Henderson, B. Gallagher, T. Eliassi-Rad, H. Tong, S. Basu, L. Akoglu, D. Koutra, C. Faloutsos, and L. Li, inProceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining, KDD ’12 (Association for Computing Machinery, New York, NY, USA, 2...

  19. [27]

    V. M. Tenorio, M. Navarro, S. Rey, S. Segarra, and A. G. Marques, in2024 58th Asilomar Conference on Signals, Systems, and Computers(2024) pp. 1080–1084, iSSN: 2576-2303

  20. [28]

    Corso, H

    G. Corso, H. Stark, S. Jegelka, T. Jaakkola, and R. Barzilay, Nature Reviews Methods Primers4, 1 (2024), publisher: Nature Publishing Group

  21. [29]

    Y.-Y. Ahn, S. E. Ahnert, J. P. Bagrow, and A.-L. Barab´ asi, Scientific Reports1, 196 (2011), publisher: Na- ture Publishing Group

  22. [30]

    Sharma, U

    T. Sharma, U. Upadhyay, J. Kalra, S. Arora, S. Ahmad, B. Aggarwal, and G. Bagler, in2020 IEEE 36th In- ternational Conference on Data Engineering Workshops (ICDEW)(2020) pp. 98–104, iSSN: 2473-3490

  23. [31]

    T. Z. Berardini, S. Mundodi, L. Reiser, E. Huala, M. Garcia-Hernandez, P. Zhang, L. A. Mueller, J. Yoon, A. Doyle, G. Lander, N. Moseyko, D. Yoo, I. Xu, B. Zoeckler, M. Montoya, N. Miller, D. Weems, and S. Y. Rhee, Plant Physiology135, 745 (2004)

  24. [32]

    Badia-i Mompel, L

    P. Badia-i Mompel, L. Wessels, S. M¨ uller-Dott, R. Trim- bour, R. O. Ramirez Flores, R. Argelaguet, and J. Saez- Rodriguez, Nature Reviews Genetics24, 739 (2023), pub- lisher: Nature Publishing Group

  25. [33]

    S. E. Ahnert and T. M. A. Fink, Journal of The Royal Society Interface13, 20160179 (2016)

  26. [34]

    Taylor-Teeples, L

    M. Taylor-Teeples, L. Lin, M. de Lucas, G. Turco, T. W. Toal, A. Gaudinier, N. F. Young, G. M. Trabucco, M. T. Veling, R. Lamothe, P. P. Handakumbura, G. Xiong, C. Wang, J. Corwin, A. Tsoukalas, L. Zhang, D. Ware, M. Pauly, D. J. Kliebenstein, K. Dehesh, I. Tagkopoulos, G. Bre...

  27. [35]

    T. Zhou, J. Ren, M. Medo, and Y.-C. Zhang, Physical Review E76, 046115 (2007), publisher: American Phys- ical Society

  28. [36]

    R. B. Mars, J. Sallet, F.-X. Neubert, and M. F. S. Rush- worth, Proceedings of the National Academy of Sciences 110, 10806 (2013), publisher: Proceedings of the Na- tional Academy of Sciences

  29. [37]

    R. B. Mars, R. E. Passingham, and S. Jbabdi, Trends in Cognitive Sciences22, 1026 (2018)

  30. [38]

    R. B. Mars, S. N. Sotiropoulos, R. E. Passingham, J. Sal- let, L. Verhagen, A. A. Khrapitchev, N. Sibson, and S. Jbabdi, eLife7, e35237 (2018), publisher: eLife Sci- ences Publications, Ltd

  31. [39]

    R. E. Passingham, K. E. Stephan, and R. K¨ otter, Nature Reviews Neuroscience3, 606 (2002), publisher: Nature Publishing Group

  32. [40]

    Bazinet, J

    V. Bazinet, J. Y. Hansen, R. Vos de Wael, B. C. Bern- hardt, M. P. van den Heuvel, and B. Misic, Nature Com- munications14, 2850 (2023), publisher: Nature Publish- 11 ing Group

  33. [41]

    Bazinet, J

    V. Bazinet, J. Y. Hansen, and B. Misic, Nature Reviews Neuroscience24, 747 (2023), publisher: Nature Publish- ing Group

  34. [42]

    Jonas and K

    E. Jonas and K. Kording, eLife4, e04250 (2015), pub- lisher: eLife Sciences Publications, Ltd

  35. [43]

    Schlegel, Y

    P. Schlegel, Y. Yin, A. S. Bates, S. Dorkenwald, K. Eich- ler, P. Brooks, D. S. Han, M. Gkantia, M. dos Santos, E. J. Munnelly, G. Badalamente, L. Serratosa Capdevila, V. A. Sane, A. M. C. Fragniere, L. Kiassat, M. W. Plei- jzier, T. St¨ urner, I. F. M. Tamimi, C. R. Dunne, I....

  36. [44]

    L. K. Scheffer, C. S. Xu, M. Januszewski, Z. Lu, S.- y. Takemura, K. J. Hayworth, G. B. Huang, K. Shi- nomiya, J. Maitlin-Shepard, S. Berg, J. Clements, P. M. Hubbard, W. T. Katz, L. Umayam, T. Zhao, D. Acker- man, T. Blakely, J. Bogovic, T. Dolafi, D. Kainmueller, T. Kawase, ...

  37. [45]

    E. C. Marin, B. J. Morris, T. St¨ urner, A. S. Cham- pion, D. Krzeminski, G. Badalamente, M. Gkantia, C. R. Dunne, K. Eichler, S.-y. Takemura, I. F. Tamimi, S. Fang, S. S. Moon, H. S. Cheong, F. Li, P. Schlegel, S. E. Ah- nert, S. Berg, J. F. P. Team, G. M. Card, M. Costa, D. ...

  38. [46]

    Haszprunar, Journal of Evolutionary Biology5, 13 (1992)

    G. Haszprunar, Journal of Evolutionary Biology5, 13 (1992)

  39. [47]

    Shepherd, R

    D. Shepherd, R. Harris, D. W. Williams, and J. W. Truman, Journal of Comparative Neurology524, 2677 (2016), publisher: John Wiley & Sons, Ltd

  40. [48]

    R. A. Jensen, Genome Biology2, interactions1002.1 (2001)

  41. [49]

    E. I. Boyle, S. Weng, J. Gollub, H. Jin, D. Botstein, J. M. Cherry, and G. Sherlock, Bioinformatics20, 3710 (2004)

  42. [50]

    Ashburner, C

    M. Ashburner, C. A. Ball, J. A. Blake, D. Botstein, H. Butler, J. M. Cherry, A. P. Davis, K. Dolinski, S. S. Dwight, J. T. Eppig, M. A. Harris, D. P. Hill, L. Issel- Tarver, A. Kasarskis, S. Lewis, J. C. Matese, J. E. Richardson, M. Ringwald, G. M. Rubin, and G. Sher- lock, Na...

  43. [51]

    H. S. Cheong, K. Eichler, T. St¨ urner, S. K. Asinof, A. S. Champion, E. C. Marin, T. B. Oram, M. Sumath- ipala, L. Venkatasubramanian, S. Namiki, I. Siwanowicz, M. Costa, S. Berg, J. F. P. Team, G. S. Jefferis, and G. M. Card, eLife13(2024), 10.7554/eLife.96084.1, pub- lisher...

  44. [52]

    Schlegel, A

    P. Schlegel, A. S. Bates, T. St¨ urner, S. R. Jagannathan, N. Drummond, J. Hsu, L. Serratosa Capdevila, A. Javier, E. C. Marin, A. Barth-Maron, I. F. Tamimi, F. Li, G. M. Rubin, S. M. Plaza, M. Costa, and G. S. X. E. Jefferis, eLife10, e66018 (2021), publisher: eLife Sciences ...

  45. [53]

    J. J. Heymans, R. E. Ulanowicz, and C. Bondavalli, Ecological Modelling149, 5 (2002). Appendix A: F ull labelled figures of connectivity vector clustering All hierarchical clustering dendrograms were created with the Euclidean metric and Ward’s method. Full reso- lution images...

  46. [54]

    Thus, aggregating the edges by incoming and outgo- ing directions groups a given organism’s prey and preda- tors respectively

    F ood web network In the food web network, the nodes represent organ- isms and the edges denote the direction of biomass trans- fer. Thus, aggregating the edges by incoming and outgo- ing directions groups a given organism’s prey and preda- tors respectively. This is the basis...

  47. [55]

    Recipe network The recipe network is a multipartite network of recipes, belonging to cuisine labels, and also associated to con- stituent ingredients. We may use a bipartite projection to build vectors of cuisines by their normalised frequency of recipes using each ingredient ...

  48. [56]

    There may be multiple GO terms associated with a single gene

    Gene regulatory network All genes in the network have associated Gene Ontol- ogy (GO) terms, that specifies their biological character- istic or role. There may be multiple GO terms associated with a single gene. A transcription factor (TF) could be regulated upstream by a TF,...

Pith tools

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