Pith. sign in

REVIEW 3 major objections 6 minor 56 references

Community and hyperedge inference in multiple hypergraphs

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read MHSBM, a multi-hypergraph stochastic block model, jointly fits hyperedges inside and between hypergraphs and the paper reports gains in community detection, hyperedge prediction, and inter-hypergraph edge inference over single-hypergraph…

desk verdict MHSBM is a legitimate extension of hypergraph SBMs to multiple linked hypergraphs, but the main evaluation leaks ground-truth labels through synthetic inter-hypergraph edges. read the letter →

arxiv 2505.04967 v1 pith:ZGEZOOPJ submitted 2025-05-08 cs.SI cs.LG

Li Ni , Ziqi Deng , Lin Mu , Lei Zhang , Wenjian Luo , Yiwen Zhang This is my paper · ORCID
classification cs.SIcs.LG
keywords multi-hypergraphstochasticblockmodelcommunitydetectionhyperedgepredictioninter-hypergraphedgeinferenceinternaldegreemixed-membershipgenerative
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 proposes MHSBM, a generative stochastic block model that fits several hypergraphs at once rather than one at a time. Its central claim is that hypergraphs linked by meaningful inter-hypergraph edges, such as the encoding relationship between genes and their protein products, contain complementary community signal, so joint inference yields more accurate communities, better prediction of missing hyperedges of any size, and better inference of missing inter-hypergraph edges. The model adds a hyperedge internal degree term, computed from how often each node appears inside sub-hyperedges, so that dominant nodes contribute more to hyperedge formation. On real high-order data the paper reports higher F1, cosine similarity, and NMI than single-hypergraph baselines, higher AUC for hyperedge prediction, and exact recovery of a known gene cluster when gene and protein hypergraphs are combined. The model also works on a single hypergraph when no inter-hypergraph edges exist.

What carries the argument

Two objects carry the argument. The first is the factorized multi-hypergraph likelihood, written as $P(A,S|\Phi)=\prod_l P_A(A^l|u^l,w^l,\theta^l)\prod_{l,l'} P_S(S^{ll'}|u^l,u^{l'},w^{ll'})$, where within-layer Poisson rates are sums over node pairs inside each hyperedge with per-node weights $\theta$, and across-layer Poisson rates are bilinear in the two membership matrices with a cross-hypergraph affinity matrix. The second is the hyperedge internal degree $\theta_{ie}=|e|\,\varepsilon_{ie}/\sum_{j\in e}\varepsilon_{je}$, which counts the fraction of sub-hyperedges of $e$ containing node $i$ and rescales that fraction by $|e|$, making node contributions to hyperedge rates unequal. EM updates derived from Jensen's inequality, together with negative-hyperedge sampling that restricts the hyperedge space to observed hyperedges and an equal number of matched-size non-hyperedges, make the likelihood tractable and produce the recovered communities and predicted edges.

What would settle it

Reconstruct the same multi-hypergraph but randomly rewire half the inter-hypergraph edges so they connect nodes from different ground-truth communities while keeping all within-hypergraph data unchanged, and check whether MHSBM's F1, NMI, and AUC remain as high as with the true edges; if they do, the claimed benefit does not come from community-aligned cross-hypergraph information.

Watch

Extended reading notes

Core claim

The central discovery the paper argues for is that the statistical evidence in one hypergraph can be sharpened by edges that connect it to another hypergraph, when those edges track the same underlying community partition. MHSBM encodes this by factorizing the joint likelihood into Poisson terms for hyperedges within each hypergraph and Poisson terms for pairwise edges between hypergraphs, all sharing node-community memberships and community affinity matrices. The named mechanism is the hyperedge internal degree, $\theta_{ie}=|e| \varepsilon_{ie}/\sum_{j\in e}\varepsilon_{je}$, where $\varepsilon_{ie}$ counts sub-hyperedges of $e$ containing node $i$; it shifts expected hyperedge rates toward nodes that dominate interaction formation. Inference is done by an expectation-maximization algorithm with variational lower bounds and negative-hyperedge sampling, and the experiments show performance degrading as inter-hypergraph edges are removed, multi-hypergraph inference beating single-hypergraph baselines on most datasets, and the gene-protein multi-hypergraph recovering a ground-truth gene cluster exactly.

Load-bearing premise

The load-bearing premise is that the inter-hypergraph edges in the experiments are generated to match the communities the model is asked to recover, so if real cross-hypergraph edges are noisy or reflect a different grouping, the reported multi-hypergraph advantage may vanish.

Editorial extensions

If this is right

  • If inter-hypergraph edges carry community-aligned signal, community detection on any single hypergraph should improve when a related hypergraph is added; the gene-protein experiment is the paper's demonstration.
  • Removing inter-hypergraph edges should degrade detection quality gradually, with the model still usable at 50% removal; the paper reports F1 staying at 0.9298 for one hypergraph.
  • Hyperedge prediction AUC should rise when multiple hypergraphs are fit jointly, with gains largest where the hypergraphs share functional relationships.
  • Inter-hypergraph edge prediction should improve as more observed inter-hypergraph edges are available; in the authorship-citation test AUC drops from 0.8056 at zero removal to roughly 0.69 when 40% or more edges are removed.
  • When no inter-hypergraph edges exist, MHSBM still functions as a single-hypergraph community detector, with the hyperedge internal degree term giving it an edge over baselines.

Reading between the lines

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

  • A natural extension is to make the model select or down-weight inter-hypergraph edges instead of treating all of them as equally informative, since the paper's own discussion notes that complementary information across hypergraphs is not always positive.
  • Replacing the pairwise coupling between hypergraphs with hyperedges that span three or more hypergraphs could capture dependencies like gene-protein-pathway chains that pairwise inter-hypergraph edges cannot represent.
  • The hyperedge internal degree is currently a deterministic function of observed sub-hyperedge containment; estimating it as a latent variable might reveal node roles that the fixed formula misses.
  • The unconstrained cross-hypergraph affinity matrix lets the model represent disassortative links between communities, and the Hospital case shows this matters; applying the same diagnostic to other datasets could identify where community structure is disassortative.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes MHSBM, a mixed-membership stochastic block model for multiple hypergraphs connected by inter-hypergraph edges. The model introduces a hyperedge internal degree to capture heterogeneous node contributions within hyperedges, and performs joint inference of communities, hyperedge presence, and inter-hypergraph edges via an EM algorithm. Experiments on real hypergraph datasets are used to claim that exploiting inter-hypergraph edges improves community detection, hyperedge prediction, and inter-hypergraph edge prediction compared with single-hypergraph baselines. A case study on a gene-protein multi-hypergraph reports exact recovery of a known gene cluster when both hypergraphs are analyzed jointly.

Significance. If the claims hold, MHSBM would be a useful extension of hypergraph stochastic block models to multi-layer and multi-domain settings, and the hyperedge internal degree is a plausible mechanism for modeling node heterogeneity in hyperedge formation. The paper ships an open-source implementation and data links, and the real gene-protein experiment is a compelling demonstration of cross-domain information flow. However, the main experimental evidence for the multi-hypergraph advantage on the benchmark datasets rests on a synthetic construction in which inter-hypergraph edges are generated from the same ground-truth communities that are later used for evaluation. This creates a potentially circular validation loop, and the derivation of the inference algorithm contains an internal inconsistency in how the hyperedge internal degree is treated. These issues concern the central claims of the paper and need to be addressed before the results can be taken as evidence of a general benefit.

major comments (3)
  1. [Section 2.2, 2.2.1, Fig. 3, Table 2] The community-detection benchmarks construct H0 and H1 by independently sampling hyperedges from a real hypergraph and then generating inter-hypergraph edges 'based on their community structures,' where those structures are the same ground-truth communities used for evaluation. Because MHSBM is the only method in Fig. 3 and Table 2 that observes the inter-hypergraph edges, its improvement over Hy-MMSBM and Hypergraph-MT may be driven by direct label information encoded in those edges rather than by a general benefit of multi-hypergraph integration. The edge-removal experiment in Fig. 2 is consistent with this interpretation: removing inter-hypergraph edges removes the label signal. I request a control experiment in which inter-hypergraph edges are degree-preservingly permuted or generated from a community-blind process, to show that the claimed advantage transfers when inter-hypergraph edges are noisy, incomplete, or only weakly aligned with communities.
  2. [Section 4.1, Eqs. (8)-(11)] The derivation of the negative term in Eq. (8) states that 'we simplify the hyperedge internal degree θ to 1,' producing Eq. (9), in which the negative term is C_l times the sum over all node pairs of u_i w u_j^T with no θ dependence. However, the positive log term in Eqs. (10) and (11) explicitly retains θ_ie θ_je inside the logarithm and in the variational distribution p. This is internally inconsistent: the model's stated mechanism for node heterogeneity appears in only one of the two terms of the objective, so the estimated parameters no longer maximize the original likelihood (2)-(3). Please either remove θ from the positive term as well, or derive an approximation of the negative term that preserves θ, and discuss the bias introduced by the current approximation.
  3. [Section 2.2] The procedure for generating inter-hypergraph edges on the synthetic benchmarks is underspecified. The text only says edges were generated 'based on their community structures' and that 'noise edges were also added,' without reporting the number of inter-hypergraph edges, the rule connecting community memberships to edge probabilities, or the noise level. This makes the multi-hypergraph experiments irreproducible and prevents a reader from judging whether the construction encodes the ground truth in a benign or a favorable way. Please provide the full generation protocol, including pseudocode or exact equations.
minor comments (6)
  1. [Section 4.1, Eq. (9)] The symbol M in the expression for C_l is not defined; please define it explicitly or remove it.
  2. [Section 2.1.3] The latent variable set Φ defined in Section 2.1.3 does not include θ, although θ appears as a model parameter in Eqs. (2)-(3); please clarify whether θ is learned during inference or computed directly from the observed hypergraph using Eq. (4).
  3. [Section 2.2.2] The caption of Fig. 3 says seven random seeds are used, while Section 2.2 says MHSBM is run 10 times; please align these numbers.
  4. [Eq. (5)] The notation PS(Sl|...) uses S^l, but the model definition in Eq. (1) uses S^{ll'}; please make the notation consistent.
  5. [Table 4] The caption mentions standard deviations, but the table shows only AUC values; either add the standard deviations or correct the caption.
  6. [General] There are several typos, including 'repectively,' 'Gene_diease,' 'hyergraph,' 'introductions,' and 'creliable'; a careful proofread is needed.

Circularity Check

1 steps flagged · score 5.0 of 10

The synthetic multi-hypergraph benchmarks inject ground-truth community labels into the inter-hypergraph edges, so MHSBM's reported multi-hypergraph advantage is partly self-constructed; the model derivation itself is not circular.

  1. self definitional [Section 2.2, first paragraph (synthetic multi-hypergraph construction)]
    "To simulate multiple hypergraph scenarios, we constructed two hypergraphs by independently sampling either 70% or 80% of hyperedges twice, labeled H0 and H1, repectively. Subsequently, inter-hypergraph edges were generated between the hypergraphs based on their community structures. Noise edges were also added."

    Equations (1) and (5) give MHSBM the inter-hypergraph edge set S as its only information beyond single-hypergraph structure. The paper states that S was generated "based on their community structures," and those are the same ground-truth communities scored by F1/NMI/CS in Fig. 3 and Table 2. MHSBM is the only method that consumes S; the baselines see only H0 and H1. Thus MHSBM's extra input is a direct function of the target labels it is asked to recover, so its advantage over single-hypergraph baselines is built into the benchmark construction rather than demonstrated as a general property of multi-hypergraph integration. A degree-preserving permutation of S or an independently noisy S would be needed to separate label leakage from genuine complementarity.

full rationale

The generative derivation of MHSBM is self-contained: Eqs. (2)-(6) extend Hy-MMSBM with an inter-hypergraph Poisson term and a data-defined hyperedge internal degree, and the EM updates in Section 4.1 are standard. The hyperedge internal degree in Eq. (4) is computed from observed sub-hyperedge counts rather than from the target communities, so it is a covariate, not a circularly defined prediction target; any overfitting concern there is a modeling risk rather than a definitional reduction. There is no load-bearing self-citation or imported uniqueness theorem. The main circularity is experimental: the synthetic community-detection evidence for the central multi-hypergraph claim is generated by constructing S from the same community labels used as ground truth. That makes MHSBM's unique input a label-bearing channel and accounts for part of the observed gain over methods that never see S. The real-data case studies provide independent support and keep the paper from being wholly circular. The Section 4.1 approximation that sets theta to 1 in the normalization term while retaining theta in the log term is an internal inconsistency that should be clarified, but it is not a circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The model introduces a new data-derived quantity (hyperedge internal degree) and relies on several distributional assumptions about hyperedge and inter-hypergraph edge generation. The key free parameters are the number of communities and the internal-degree simplification used in the approximate inference objective.

free parameters (4)
  • Number of communities K per hypergraph = Set to ground-truth community count (Table 1)
    The model requires K as input; in all experiments K is taken from known ground truth, which is an oracle quantity unavailable in real unsupervised use.
  • Negative hyperedge sampling size = |E^{l-}| = |E^{l+}|
    The set of non-observed hyperedges is sampled with the same size as observed, a heuristic inherited from prior work but not separately justified.
  • Theta simplification in negative term = theta = 1
    In the derivation of the objective, the hyperedge internal degree is set to 1 in the approximation of the sum over all hyperedges (Methods, Eq. 9), creating a mismatch with the generative model.
  • Assortative initialization flag = True or False chosen per dataset
    The initialization of the community affinity matrix is an experimental hyperparameter; performance varies substantially with it (Fig. 4b), so results depend on this choice.
assumptions (5)
  • domain assumption Hyperedge weights are independent Poisson variables given latent memberships, affinities, and internal degrees (Eq. 2).
    Used in Eq. (2); a standard SBM assumption but not tested against alternative noise models.
  • domain assumption Inter-hypergraph edges are conditionally independent Poisson variables with rate u_i w^{ll'} u_j^T (Eq. 5-6).
    Modeling assumption for inter-hypergraph edges; no empirical validation that such edges follow this form.
  • domain assumption Hypergraphs and inter-hypergraph edges are conditionally independent given latent variables (Eq. 1).
    Factorized likelihood in Eq. (1) assumes no other dependencies.
  • standard math Jensen's inequality lower bound and coordinate ascent updates converge to a local optimum of the surrogate objective.
    Standard EM approach; the paper acknowledges only a local optimum is guaranteed.
  • ad hoc to paper The sum over all possible hyperedges can be approximated by the average interaction strength times the number of node pairs (Methods, Eq. 9).
    This computationally motivated approximation also sets theta to 1; it is not derived from the generative model and its accuracy is not evaluated.
invented entities (1)
  • Hyperedge internal degree theta_ie
    purpose: Quantifies each node's contribution to hyperedge formation as |e| * epsilon_ie / sum epsilon_je (Eq. 4).
    Constructed from the observed hypergraph via sub-hyperedge counts; it is an ad hoc modeling construct rather than an independently measured quantity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Community and hyperedge inference in multiple hypergraphs." pith.science (2026). https://pith.science/paper/ZGEZOOPJ

@misc{pith2026250504967,
  author       = {Pith},
  title        = {Pith review of: Community and hyperedge inference in multiple hypergraphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZGEZOOPJ}},
  note         = {Machine review of arXiv:2505.04967}
}
read the original abstract

Hypergraphs, capable of representing high-order interactions via hyperedges, have become a powerful tool for modeling real-world biological and social systems. Inherent relationships within these real-world systems, such as the encoding relationship between genes and their protein products, drive the establishment of interconnections between multiple hypergraphs. Here, we demonstrate how to utilize those interconnections between multiple hypergraphs to synthesize integrated information from multiple higher-order systems, thereby enhancing understanding of underlying structures. We propose a model based on the stochastic block model, which integrates information from multiple hypergraphs to reveal latent high-order structures. Real-world hyperedges exhibit preferential attachment, where certain nodes dominate hyperedge formation. To characterize this phenomenon, our model introduces hyperedge internal degree to quantify nodes' contributions to hyperedge formation. This model is capable of mining communities, predicting missing hyperedges of arbitrary sizes within hypergraphs, and inferring inter-hypergraph edges between hypergraphs. We apply our model to high-order datasets to evaluate its performance. Experimental results demonstrate strong performance of our model in community detection, hyperedge prediction, and inter-hypergraph edge prediction tasks. Moreover, we show that our model enables analysis of multiple hypergraphs of different types and supports the analysis of a single hypergraph in the absence of inter-hypergraph edges. Our work provides a practical and flexible tool for analyzing multiple hypergraphs, greatly advancing the understanding of the organization in real-world high-order systems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 55 canonical work pages

  1. [1]

    & Vaccarino, F

    Patania, A., Petri, G. & Vaccarino, F. The shape of collaborations.EPJ Data Science6, 1–16 (2017)

    work page 2017

  2. [2]

    & Theis, F

    Klamt, S., Haus, U.-U. & Theis, F. Hyper- graphs and cellular networks.PLOS Compu- tational Biology5, 1–6 (2009)

    work page 2009

  3. [3]

    Zimmer, A., Katzir, I., Dekel, E., Mayo, A. E. & Alon, U. Prediction of multi- dimensional drug dose responses based on measurements of drug pairs.Proceedings of the National Academy of Sciences113, 10442–10447 (2016)

    work page 2016

  4. [4]

    Internationalmachine learning society (imls).International Confer- ence on Machine Learning35(2018)

    Li,P.&Milenkovic,O. Internationalmachine learning society (imls).International Confer- ence on Machine Learning35(2018)

    work page 2018

  5. [5]

    F., Petri, G

    De Arruda, G. F., Petri, G. & Moreno, Y. Social contagion models on hypergraphs. Physical Review Research2, 023032 (2020)

    work page 2020

  6. [6]

    & Lam- biotte, R

    Sahasrabuddhe, R., Neuhäuser, L. & Lam- biotte, R. Modelling non-linear consen- sus dynamics on hypergraphs.Journal of Physics: Complexity2, 025006 (2021)

    work page 2021

  7. [7]

    Physics Reports874, 1–92 (2020)

    Battiston, F.et al.Networks beyond pair- wise interactions: Structure and dynamics. Physics Reports874, 1–92 (2020)

    work page 2020

  8. [8]

    S., Bassett, D

    Torres, L., Blevins, A. S., Bassett, D. & Eliassi-Rad, T. The why, how, and when of 11 representations for complex systems.SIAM Review63, 435–485 (2021)

    work page 2021

Show all 56 references
  1. [9]

    & Petri, G.Higher-Order Systems(Springer International Publishing, 2022)

    Battiston, F. & Petri, G.Higher-Order Systems(Springer International Publishing, 2022)

    2022

  2. [10]

    Benson, A. R. Three hypergraph eigenvector centralities.SIAM Journal on Mathematics of Data Science1, 293–312 (2019)

    2019

  3. [11]

    & Higham, D

    Tudisco, F. & Higham, D. J. Node and edge nonlinear eigenvector centrality for hyper- graphs.Communications Physics4, 201 (2021)

    2021

  4. [12]

    & De Bacco, C

    Contisciani, M., Battiston, F. & De Bacco, C. Inference of hyperedges and overlap- ping communities in hypergraphs.Nature communications13, 7229 (2022)

    2022

  5. [13]

    R., Abebe, R., Schaub, M

    Benson, A. R., Abebe, R., Schaub, M. T., Jadbabaie, A. & Kleinberg, J. Simplicial closure and higher-order link prediction.Pro- ceedings of the National Academy of Sciences 115, E11221–E11230 (2018)

    2018

  6. [14]

    & Schwartz, O

    Ballard, G., Druinsky, A., Knight, N. & Schwartz, O. Hypergraph partition- ing for sparse matrix-matrix multiplication. Acm Transactions On Parallel Computing3 (2016)

    2016

  7. [15]

    & Shalita, A

    Kabiljo, I., Karrer, B., Pundir, M., Pupyrev, S. & Shalita, A. Social hash partitioner: a scalable distributed hypergraph partitioner. Proceedings Of The Vldb Endowment10, 1418–1429 (2017)

    2017

  8. [16]

    & Shekhar, S

    Karypis, G., Aggarwal, R., Kumar, V. & Shekhar, S. Multilevel hypergraph parti- tioning: applications in vlsi domain.IEEE Transactions on Very Large Scale Integration Systems7, 69–79 (1999)

    1999

  9. [17]

    2005 IEEE Conference on Computer Vision and Pattern Recognition2, 838–845 (2005)

    Agarwal, S.et al.Beyond pairwise clustering. 2005 IEEE Conference on Computer Vision and Pattern Recognition2, 838–845 (2005)

    2005

  10. [18]

    T., Shi, F

    Ke, Z. T., Shi, F. & Xia, D. Commu- nity detection for hypergraph networks via regularized tensor power iteration (2020). Preprint at https://arxiv.org/abs/1909. 06503, arXiv:1909.06503

    2020 arXiv

  11. [19]

    Zhong, H.et al.Penalized flow hyper- graph local clustering.IEEE Transactions on Knowledge and Data Engineering36, 2110– 2125 (2024)

    2024

  12. [20]

    Howchoosing random-walk model and network represen- tation matters for flow-based community detection in hypergraphs.Communications Physics4, 133 (2021)

    Eriksson, A., Edler, D., Rojas, A., deDomenico,M.&Rosvall,M. Howchoosing random-walk model and network represen- tation matters for flow-based community detection in hypergraphs.Communications Physics4, 133 (2021)

    2021

  13. [21]

    & Schölkopf, B

    Zhou, D., Huang, J. & Schölkopf, B. Learning with hypergraphs: Clustering, classification, and embedding.Advances in Neural Infor- mation Processing Systems19, 1601–1608 (2006)

    2006

  14. [22]

    & Dukkipati, A

    Ghoshdastidar, D. & Dukkipati, A. A prov- able generalized tensor spectral method for uniform hypergraph partitioning.Proceed- ings of the 32nd International Conference on Machine Learning32, 400–409 (2015)

    2015

  15. [23]

    C., Caltagirone, F., Krząkała, F

    Angelini, M. C., Caltagirone, F., Krząkała, F. & Zdeborová, L. Spectral detection on sparse hypergraphs.2015 Annual Allerton Conference on Communication, Control, and Computing53, 66–73 (2015)

    2015

  16. [24]

    S., Veldt, N

    Chodrow, P. S., Veldt, N. & Benson, A. R. Generative hypergraph clustering: From blockmodels to modularity.Science Advances7, eabh1303 (2021)

    2021

  17. [25]

    & Bacco, C

    Ruggeri, N., Contisciani, M., Battiston, F. & Bacco, C. D. Community detection in large hypergraphs.Science Advances9, eadg9159 (2023)

    2023

  18. [26]

    & De Bacco, C

    Badalyan, A., Ruggeri, N. & De Bacco, C. Structure and inference in hypergraphs with node attributes.Nature Communications15, 7073 (2024)

    2024

  19. [27]

    & Lambiotte, R

    Carletti, T., Fanelli, D. & Lambiotte, R. Random walks and community detection in hypergraphs.Journal of Physics: Complexity 2, 015011 (2021)

    2021

  20. [28]

    & Gilani, S

    Zhang, E., Suter, D., Truong, G. & Gilani, S. Z. Sparse hypergraph community detec- tion thresholds in stochastic block model. Advances in Neural Information Processing Systems35, 34012–34023 (2022)

    2022

  21. [29]

    & Sandon, C

    Abbe, E. & Sandon, C. Recovering commu- nities in the general stochastic block model without knowing the parameters.Advances in Neural Information Processing Systems 28, 676–684 (2015)

    2015

  22. [30]

    & De Bacco, C

    Ruggeri, N., Battiston, F. & De Bacco, C. Framework to generate hypergraphs with community structure.Physical Review E 109, 034309 (2024)

    2024

  23. [31]

    & Busiello, D

    Nicoletti, G. & Busiello, D. M. Mutual infor- mation disentangles interactions from chang- ing environments.Physical Review Letters 127, 228301 (2021)

    2021

  24. [32]

    & Cai, D

    Zhou, D., Xiao, Y., Zhang, Y., Xu, Z. & Cai, D. Granger causality network reconstruc- tion of conductance-based integrate-and-fire 12 neuronal systems.Plos one9, e87636 (2014)

    2014

  25. [33]

    & Droit, A

    Bodein, A., Scott-Boyer, M.-P., Perin, O., Lê Cao, K.-A. & Droit, A. Interpretation of network-based integration from multi-omics longitudinal data.Nucleic acids research50, e27–e27 (2022)

    2022

  26. [34]

    elegans.Nature640, 194–202 (2025)

    Zhang, H.et al.A systems-level, semi- quantitative landscape of metabolic flux in c. elegans.Nature640, 194–202 (2025)

    2025

  27. [35]

    B.et al.Entire expressed periph- eral blood transcriptome in pediatric severe malarial anemia.Nature Communications 15, 5037 (2024)

    Anyona, S. B.et al.Entire expressed periph- eral blood transcriptome in pediatric severe malarial anemia.Nature Communications 15, 5037 (2024)

    2024

  28. [36]

    & Loscalzo, J

    Barabási, A.-L., Gulbahce, N. & Loscalzo, J. Network medicine: a network-based approach to human disease.Nature reviews genetics 12, 56–68 (2011)

    2011

  29. [37]

    Zhao, Z.et al.Multi-omics integration anal- ysis unveils heterogeneity in breast cancer at theindividuallevel.Cell Cycle22,2229–2244 (2023)

    2023

  30. [38]

    & Albert, R

    Barabási, A.-L. & Albert, R. Emergence of scaling in random networks.Science286, 509–512 (1999)

    1999

  31. [39]

    ACM63, 139–144 (2020)

    Goodfellow, I.et al.Generative adversar- ial networks.Commun. ACM63, 139–144 (2020)

    2020

  32. [40]

    Bianconi, G.Multilayer Networks: Struc- ture and Function(Oxford University Press, 2018)

    2018

  33. [41]

    & Moreno, Y

    Aleta, A. & Moreno, Y. Multilayer networks in a nutshell.Annual Review of Condensed Matter Physics10, 45–62 (2019)

    2019

  34. [42]

    P., Laird, N

    Dempster, A. P., Laird, N. M. & Rubin, D. B. Maximum likelihood from incomplete data via the em algorithm.Journal of the Royal Statistical Society. Series B (Methodological) 39, 1–38 (1977)

    1977

  35. [43]

    & Barrat, A

    Mastrandrea, R., Fournet, J. & Barrat, A. Contact patterns in a high school: A compar- ison between data collected using wearable sensors, contact diaries and friendship sur- veys.Plos one10, e0136497 (2015)

    2015

  36. [44]

    Stehlé, J.et al.High-resolution measure- ments of face-to-face contact patterns in a primary school.Plos one6, e23176 (2011)

    2011

  37. [45]

    GÉNOIS, M.et al.Data on face-to-face con- tacts in an office building suggest a low-cost vaccination strategy based on community linkers.Network Science3, 326–347 (2015)

    2015

  38. [46]

    Vanhems, P.et al.Estimating potential infec- tion transmission routes in hospital wards using wearable proximity sensors.Plos one 8, 1–9 (2013)

    2013

  39. [47]

    Fowler, J. H. Connecting the congress: A study of cosponsorship networks.Political Analysis14, 456–487 (2006)

    2006

  40. [48]

    Mets, A. A philosophical critique of the distinction of representational and pragmatic measurements on the example of the periodic system of chemical elements.Foundations of Science24, 73–93 (2019)

    2019

  41. [49]

    Liu, X.et al.Robustness and lethality in mul- tilayer biological molecular networks.Nature communications11, 6043 (2020)

    2020

  42. [50]

    Pedregosa, F.et al.Scikit-learn: Machine learning in python12, 2825–2830 (2011)

    2011

  43. [51]

    Muc-4 evaluation metrics.Pro- ceedings of the 4th Conference on Message Understanding4, 22–29 (1992)

    Chinchor, N. Muc-4 evaluation metrics.Pro- ceedings of the 4th Conference on Message Understanding4, 22–29 (1992)

    1992

  44. [52]

    & Ghosh, J

    Strehl, A. & Ghosh, J. Cluster ensembles — a knowledge reuse framework for combin- ing multiple partitions.Journal of Machine Learning Research3, 583–617 (2003)

    2003

  45. [53]

    & Are- nas, A

    Danon, L., Díaz-Guilera, A., Duch, J. & Are- nas, A. Comparing community structure identification.Journal of Statistical Mechan- ics: Theory and Experiment2005, P09008 (2005)

    2005

  46. [54]

    Information theoretic measures for clusterings compari- son: Variants, properties, normalization and correction for chance.Journal of Machine Learning Research11, 2837–2854 (2010)

    Vinh,N.X.,Epps,J.&Bailey,J. Information theoretic measures for clusterings compari- son: Variants, properties, normalization and correction for chance.Journal of Machine Learning Research11, 2837–2854 (2010)

    2010

  47. [55]

    Shannon, C. E. A mathematical theory of communication.Mobile Computing and Communications Review5, 3–55 (2001)

    2001

  48. [56]

    Do, C. B. & Batzoglou, S. What is the expectation maximization algorithm?Nature Biotechnology26, 897–899 (2008). Acknowledgements This work was supported by the National Nat- ural Science Foundation of China [No.62106004, No.62272001, and No.62206004]. Author contributions L.N...

    2008

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.