Pith. sign in

REVIEW 3 major objections 5 minor 60 references

Bayesian DAG Structure Learning with Simultaneous Shrinkage Covariance Estimation under Scale-Mixture Error Distributions in the Proportional High-Dimensional Regime

T0 review · 3 major / 5 minor · reviewed 2026-07-10 · grok-4.5

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

Pith's one-line read A single Bayesian prior recovers directed gene networks and precision matrices at high dimension even when errors are heavy-tailed.

desk verdict Solid joint construction of horseshoe-Cholesky DAG + scale-mixture robustness with useful rates, but the central concentration lemma is only sketched and the TCGA application is synthetic. read the letter →

arxiv 2607.08428 v1 pith:CL4GT3CY submitted 2026-07-09 stat.ME

Samaneh Nazari , Mohammad Arashi , Abdolnasser Sadeghkhani This is my paper
classification stat.ME MSC 62F1562H2262J0762F1262G2062P10
keywords BayesianDAGlearninghorseshoepriormodifiedCholeskydecompositionscale-mixtureofnormalsposteriorcontractionhigh-dimensionalasymptoticsRNA-seqrobustprecisionestimation
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 you can learn both a directed acyclic graph and its precision matrix at once, even when the number of variables is a fixed fraction of the sample size and the observations are heavy-tailed or contaminated. The method places a horseshoe shrinkage prior on the free entries of the modified Cholesky factor of the precision matrix; zeros in that factor become missing parent edges once a topological order is chosen. A latent scale for each observation automatically down-weights outliers and yields Student-t, Laplace or slash margins as special cases. The authors prove that the joint posterior contracts at the rate sqrt((s0 log p)/n) and that the recovered edge set matches the truth with high probability under a beta-min condition. Simulations and a TCGA-calibrated RNA-seq panel show that the procedure keeps false-positive edges far lower than Gaussian competitors while still recovering known master regulators.

What carries the argument

The R-DACH prior: a global-local horseshoe prior on the strictly lower-triangular entries of the modified Cholesky factor L of the precision matrix Ω = Lᵀ D⁻¹ L, together with per-observation inverse-gamma latent scales. Sparsity in L simultaneously encodes parent-set selection and continuous shrinkage of edge strengths.

What would settle it

Generate data from a known sparse DAG whose smallest nonzero Cholesky entry is deliberately set just below C0 sqrt((log p)/n); if the posterior modal skeleton still recovers the true edge set with high probability as n grows, the claimed selection consistency is false.

Watch

Extended reading notes

Core claim

Under the proportional regime p/n o c and scale-mixture errors, the R-DACH posterior contracts jointly for the Cholesky factor and the precision matrix at rate ε_n = M sqrt((s0 log p)/n), and the posterior modal skeleton recovers the true edge set with probability tending to 1 when every true nonzero Cholesky entry exceeds C0 sqrt((log p)/n).

Load-bearing premise

Every true nonzero edge must be at least a constant times sqrt((log p)/n) large; if any true edge is weaker than that, the horseshoe rule cannot reliably separate it from noise and skeleton consistency fails.

Editorial extensions

If this is right

  • In high-dimensional RNA-seq panels, regulatory networks can be recovered with far fewer false-positive edges under residual batch contamination than Gaussian Cholesky or graphical-horseshoe baselines.
  • The same contraction rate holds (up to a log-log factor) when the observations possess only 2+δ moments, so the method remains valid for overdispersed count data after mild transformation.
  • Per-iteration cost scales as O(np d-bar) for sparse graphs, making joint order-and-structure MCMC feasible for p of several hundred.
  • Master-regulator hubs remain identifiable while background spurious edges are suppressed, improving the interpretability of downstream pathway analyses.

Reading between the lines

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

  • The same Cholesky-horseshoe construction could be fused across time to track smoothly evolving regulatory networks without re-deriving the entire contraction theory.
  • Because the latent scales already down-weight outliers, replacing the Gaussian copula margins with discrete or zero-inflated margins would extend the method to raw count data while preserving the DAG-Markov property.
  • The β-min threshold suggests a practical pre-screen: edges whose absolute partial correlation falls near the noise floor should be excluded from the active set before running the Gibbs sampler.
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

3 major / 5 minor

Summary. The paper introduces R-DACH, a Bayesian hierarchical model that places a horseshoe prior on the strictly lower-triangular entries of the modified Cholesky factor of a DAG-Markov precision matrix, together with a per-observation inverse-gamma scale mixture for heavy-tailed robustness and a uniform prior on topological orderings. A partially-collapsed blocked Gibbs sampler is given. Under assumptions (A1)–(A5) and (A7) the authors claim joint posterior contraction of the Cholesky factor L (Frobenius) and precision Ω (operator norm) at rate ε_n = M √((s_{0,n} log p_n)/n) in the proportional regime p_n/n → c ∈ (0,∞) (Theorem 3.4), skeleton selection consistency under an additional β-min condition (Theorem 3.5), and rate preservation (up to √(log log n)) under 2+δ moments (Theorem 3.6). Simulations and a TCGA-calibrated RNA-seq example are used to illustrate gains under contamination.

Significance. If the contraction and selection results hold rigorously, the work fills a genuine gap: a single coherent prior that couples DAG structure with continuous global–local shrinkage, admits heavy tails, and is analysed in the practically relevant proportional regime. The computational complexity claim O(np d̄ + p d̄²) for sparse graphs and the public R/Python implementations are concrete strengths. The RNA-seq application, while synthetic, is carefully calibrated and shows a clear false-positive reduction that is of practical interest for regulatory-network inference. These contributions would be of interest to the high-dimensional graphical-model and Bayesian nonparametrics communities.

major comments (3)
  1. Theorem 3.4 is the central claim. Its proof relies on Lemma 3.1, which asserts that the weighted sample covariance bS_n = n^{-1} ∑ ω_i y_i y_i^⊤ concentrates in operator norm at the usual sub-Gaussian rate after the random scales ω_i are integrated out. The argument given is only that “conditional on ω the Wishart bound holds and integration multiplies by a constant.” In the proportional regime the largest eigenvalue already sits at the Marchenko–Pastur edge; random multiplicative weights whose second moments are merely finite (A5) can inflate the operator-norm deviation by a factor that depends on the tails of π_ω. Without a uniform-in-ω or moment-generating-function argument that controls this inflation, the exponential tests required by the Ghosal–van der Vaart template are not guaranteed, and the claimed rate may fail. The same gap reappears (with a log-log factor) in Theorem 3.6. A
  2. The simulation design (Section 4) and the RNA-seq experiment (Section 5) compare R-DACH only against its own Gaussian-CH ablation (and, in the abstract, against methods that are never tabulated). The abstract claims “substantial gains over graphical-horseshoe, DAG-Wishart, and PC-based competitors,” yet Tables 1–3 and Figures 2–3 contain no such competitors. Without those baselines the empirical support for the practical advantage of the joint construction remains incomplete.
  3. Assumption (A6) (β-min) is load-bearing for skeleton consistency (Theorem 3.5). The constant C_0 is left unspecified and the paper itself notes that performance degrades when the bound is approached. A short calibration study or a data-driven diagnostic for when the β-min condition is plausible would strengthen the claim that the selection result is usable in practice.
minor comments (5)
  1. The abstract and introduction list graphical-horseshoe, DAG-Wishart and PC as competitors, but the simulation section never reports them; either add the comparisons or soften the abstract claim.
  2. Figure 1 caption and axis labels use “s0log p/n” without the square-root symbol that appears in the text; the linear-fit slope of 1.26 is therefore hard to interpret at a glance.
  3. Hyper-parameter defaults (α_d = β_d = 10^{-3}, τ_0 = 1/(p √(n log p)), edge threshold 0.30 √((log p)/n)) are stated without sensitivity checks; a brief paragraph or supplement table would help reproducibility.
  4. The ordering prior is described as uniform, yet moves are restricted to adjacent transpositions; the resulting stationary distribution on S_p should be stated explicitly.
  5. Typographical inconsistencies appear in the arXiv header date (July 10, 2026) and in the repeated use of “bSn” versus “bS_n”.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: contraction rates and selection consistency follow from standard Ghosal–van der Vaart machinery applied to a new horseshoe-Cholesky prior; no step reduces by construction to a fitted input or self-citation.

full rationale

The central claims (Theorems 3.4–3.6) are obtained by verifying the three ingredients of the Ghosal–van der Vaart posterior-contraction template (prior mass in a KL ball via the horseshoe lower bound of Lemma 3.2, a sieve of controlled support size, and exponential tests built from the matrix-concentration inequality of Lemma 3.1) under the stated assumptions (A1)–(A7). These ingredients are independent of any data-dependent fit: the global-scale hyperparameter τ0 is taken from the external high-dimensional horseshoe literature (van der Pas et al., 2014), the diagonal IG hyperparameters are fixed at conventional weak values 10^{-3}, and the rate ε_n = M √((s0 log p)/n) is the generic sparse rate, not a quantity estimated from the simulations or the TCGA panel. The application uses a synthetic DAG whose moments are calibrated to TCGA only for reproducibility; recovery metrics are reported against the known planted edge set, so no “prediction” is forced by the calibration. No uniqueness theorem or ansatz is imported from the authors’ own prior work, and the bibliography contains no self-citations that carry load-bearing weight. Minor gaps in the justification of Lemma 3.1 (integration over random scales) affect correctness, not circularity. The derivation chain is therefore self-contained against external benchmarks.

Assumptions & free parameters 4 free parameters · 7 assumptions · 1 invented entities

The central contraction and selection claims rest on seven explicit high-dimensional assumptions (A1)–(A7), standard concentration tools for Wishart and Fuk–Nagaev matrix sums, and the classical Ghosal–van der Vaart posterior-contraction theorem. Free parameters are the usual weakly-informative hyperparameters of the horseshoe and inverse-gamma priors plus the β-min constant C0. No new physical entities are postulated; the R-DACH prior itself is an invented modelling construct whose independent evidence is the simulation and application performance.

free parameters (4)
  • τ0 (global horseshoe scale) = 1/(p √n log p)
    Set by hand to 1/(p √n log p) following van der Pas et al. (2014); the contraction rate depends on this choice (A7).
  • α_d, β_d (diagonal IG hyperparameters) = 10^{-3}
    Defaulted to 10^{-3}; weakly informative but still free choices that affect finite-sample behaviour of D.
  • C0 (β-min constant) = large enough
    Required large enough for skeleton consistency (A6); not estimated from data and controls the separation of true edges from noise.
  • edge-selection threshold 0.30 √((log p)/n) = 0.30
    Used in simulations for declaring edges; a free numerical factor multiplying the theoretical rate.
assumptions (7)
  • domain assumption p_n / n → c ∈ (0,∞) (proportional regime)
    Assumption (A1); the entire asymptotic analysis is conducted under this regime rather than the classical p = o(n) setting.
  • domain assumption Uniform eigenvalue bounds κ ≤ λ_min(Ω0) ≤ λ_max(Ω0) ≤ κ̄
    Assumption (A2); needed for equivalence of Hellinger, Frobenius and operator norms and for concentration of the weighted sample covariance.
  • domain assumption s0,n log p_n = o(n)
    Assumption (A3); sparsity condition that keeps the prior-mass and entropy terms smaller than the likelihood.
  • domain assumption β-min: min |L0,jk| ≥ C0 √((log p)/n) for true edges
    Assumption (A6); load-bearing for skeleton consistency (Theorem 3.5).
  • standard math Ghosal–van der Vaart posterior contraction theorem for non-i.i.d. models
    Invoked as the general machinery that converts prior mass + tests + sieve into contraction (proof of Theorem 3.4).
  • standard math Wishart / Fuk–Nagaev matrix concentration bounds
    Lemma 3.1 and its heavy-tail extension rely on Vershynin-type concentration for sums of independent random matrices.
  • ad hoc to paper Horseshoe prior mass lower bound on sparse balls
    Lemma 3.2 supplies a paper-specific lower bound on the horseshoe measure of an ε-ball around a sparse L0; the constants depend on the particular auxiliary IG representation chosen.
invented entities (1)
  • R-DACH prior (horseshoe on Cholesky factor + per-observation IG scale mixture + uniform ordering prior)
    purpose: Simultaneously encode DAG-Markov structure, continuous shrinkage and heavy-tail robustness inside a single hierarchical model amenable to Gibbs sampling and contraction analysis.
    The combination is new; its independent evidence is limited to the simulations and the synthetic TCGA panel reported in the paper itself.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bayesian DAG Structure Learning with Simultaneous Shrinkage Covariance Estimation under Scale-Mixture Error Distributions in the Proportional High-Dimensional Regime." pith.science (2026). https://pith.science/paper/CL4GT3CY

@misc{pith2026260708428,
  author       = {Pith},
  title        = {Pith review of: Bayesian DAG Structure Learning with Simultaneous Shrinkage Covariance Estimation under Scale-Mixture Error Distributions in the Proportional High-Dimensional Regime},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CL4GT3CY}},
  note         = {Machine review of arXiv:2607.08428}
}
abstract

We propose a unified Bayesian framework namely robust DAG-Cholesky horseshoe (R-DACH) for joint directed acyclic graph (DAG) structure learning and precision matrix estimation in the high-dimensional proportional asymptotic regime $p/n \to c \in (0,\infty)$, under the scale mixture of normal errors. The construction places a global-local horseshoe-type prior directly on the strictly lower-triangular entries of the modified Cholesky factor of the DAG-Markov precision matrix, so that sparsity in the Cholesky parameters induces a coherent parent-set selection consistent with a topological ordering of the variables. A per-observation inverse-gamma scale mixture yields automatic robustness to heavy-tailed and contaminated observations and admits Student-$t$, Laplace, and slash distributions as special cases. We design a partially-collapsed blocked Gibbs sampler that traverses the joint space of orderings, sparsity patterns and continuous parameters. Simulations across $(n,p)$ configurations with $p$ up to several hundreds confirm the theoretical rates and demonstrate substantial gains over graphical-horseshoe, DAG-Wishart, and PC-based competitors under contamination. An application to RNA-seq gene-expression data from \emph{The Cancer Genome Atlas} reveals biologically interpretable regulatory structure that competing methods fail to recover.

Figures

Figures reproduced from arXiv: 2607.08428 by the authors.

Figure 1
Figure 1. Empirical posterior-mean error of Lˆ (circles) and Ωˆ (squares) against the theoretical rate p s0 log p/n, p = 40, t4 errors, n ∈ {80, 120, 200, 300, 500}. Linear fit shown as dashed grey. 4.3 Selection accuracy and operator-norm performance 30 40 50 60 70 80 90 100 p 0.0 0.2 0.4 0.6 0.8 1.0 F1 score errors: gaussian 30 40 50 60 70 80 90 100 p 0.0 0.2 0.4 0.6 0.8 1.0 F1 score errors: t R-DACH, =0% R-DACH, =10% Gauss… view at source ↗
Figure 2
Figure 2. F1 score for DAG-skeleton recovery as a function of [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Operator-norm error ∥Ωˆ − Ω0∥op as a function of p for n = 200. R-DACH error is bounded under all conditions; Gaussian-CH error grows under heavy tails and contamination. when contamination is present [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: R-DACH wall-clock time per 300-iteration run on a single CPU core. The cost is [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: R-DACH posterior edge-inclusion probability matrix (left) and the calibrated ground [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: In-degree distribution of the estimated DAG skeletons under R-DACH and Gauss-CH. [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages

  1. [1]

    Altomare, D., Consonni, G., & La Rocca, L. (2013). Objective Bayesian search of Gaussian directed acyclic graphical models for ordered variables with non-local priors. Biometrics, 69(2), 478--487

    work page 2013

  2. [2]

    F., & Mallows, C

    Andrews, D. F., & Mallows, C. L. (1974). Scale mixtures of normal distributions. Journal of the Royal Statistical Society. Series B (Methodological), 36(1), 99--102

    work page 1974

  3. [3]

    A., et al

    Badve, S., Turbin, D., Thorat, M. A., et al. (2007). FOXA1 expression in breast cancer -- correlation with luminal subtype A and survival. Clinical Cancer Research, 13(15), 4415--4421

    work page 2007

  4. [4]

    Bai, Z., & Silverstein, J. W. (2010). Spectral Analysis of Large Dimensional Random Matrices, 2nd edition. New York: Springer

    work page 2010

  5. [5]

    Banerjee, S., & Ghosal, S. (2014). Posterior convergence rates for estimating large precision matrices using graphical models. Electronic Journal of Statistics, 8(2), 2111--2137

    work page 2014

  6. [6]

    Barab \'a si, A.-L., & Oltvai, Z. N. (2004). Network biology: understanding the cell's functional organization. Nature Reviews Genetics, 5(2), 101--113

    work page 2004

  7. [7]

    M., & Berger, J

    Barbieri, M. M., & Berger, J. O. (2004). Optimal predictive model selection. The Annals of Statistics, 32(3), 870--897

    work page 2004

  8. [8]

    A., Stolovitzky, G., Klein, U., Dalla-Favera, R., & Califano, A

    Basso, K., Margolin, A. A., Stolovitzky, G., Klein, U., Dalla-Favera, R., & Califano, A. (2005). Reverse engineering of regulatory networks in human B cells. Nature Genetics, 37(4), 382--390

    work page 2005

Show all 60 references
  1. [9]

    Bello, K., Aragam, B., & Ravikumar, P. (2022). DAGMA: Learning DAGs via M-matrices and a log-determinant acyclicity characterization. Advances in Neural Information Processing Systems, 35, 8226--8239

    2022

  2. [10]

    Ben-David, E., Li, T., Massam, H., & Rajaratnam, B. (2011). High dimensional Bayesian inference for Gaussian directed acyclic graph models. arXiv preprint arXiv:1109.4371

    2011 arXiv

  3. [11]

    M., Kucukelbir, A., & McAuliffe, J

    Blei, D. M., Kucukelbir, A., & McAuliffe, J. D. (2017). Variational inference: a review for statisticians. Journal of the American Statistical Association, 112(518), 859--877

    2017

  4. [12]

    Cao, X., Khare, K., & Ghosh, M. (2019). Posterior graph selection and estimation consistency for high-dimensional Bayesian DAG models. The Annals of Statistics, 47(1), 319--348

    2019

  5. [13]

    M., Polson, N

    Carvalho, C. M., Polson, N. G., & Scott, J. G. (2010). The horseshoe estimator for sparse signals. Biometrika, 97(2), 465--480

    2010

  6. [14]

    L., & Peluso, S

    Castelletti, F., Consonni, G., Della Vedova, M. L., & Peluso, S. (2018). Learning Markov equivalence classes of directed acyclic graphs: an objective Bayes approach. Bayesian Analysis, 13(4), 1235--1260

    2018

  7. [15]

    Castelo, R., & Roverato, A. (2009). Reverse engineering molecular regulatory networks from microarray data with qp-graphs. Journal of Computational Biology, 16(2), 213--227

    2009

  8. [16]

    C., Olsen, C., et al

    Colaprico, A., Silva, T. C., Olsen, C., et al. (2016). TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. Nucleic Acids Research, 44(8), e71

    2016

  9. [17]

    G., Dawid, A

    Cowell, R. G., Dawid, A. P., Lauritzen, S. L., & Spiegelhalter, D. J. (1999). Probabilistic Networks and Expert Systems. New York: Springer

    1999

  10. [18]

    Cui, R., Groot, P., & Heskes, T. (2016). Copula PC algorithm for causal discovery from mixed data. In Machine Learning and Knowledge Discovery in Databases (ECML PKDD 2016), pp. 377--392. Springer

    2016

  11. [19]

    Fern \'a ndez, C., & Steel, M. F. J. (1999). Multivariate Student- t regression models: pitfalls and inference. Biometrika, 86(1), 153--167

    1999

  12. [20]

    Finegold, M., & Drton, M. (2011). Robust graphical modeling of gene networks using classical and alternative t -distributions. The Annals of Applied Statistics, 5(2A), 1057--1080

    2011

  13. [21]

    Friedman, N., & Koller, D. (2003). Being Bayesian about network structure. A Bayesian approach to structure discovery in Bayesian networks. Machine Learning, 50(1--2), 95--125

    2003

  14. [22]

    Friedman, N. (2004). Inferring cellular networks using probabilistic graphical models. Science, 303(5659), 799--805

    2004

  15. [23]

    Geiger, D., & Heckerman, D. (2002). Parameter priors for directed acyclic graphical models and the characterization of several probability distributions. The Annals of Statistics, 30(5), 1412--1440

    2002

  16. [24]

    Ghosal, S., & van der Vaart, A. W. (2007). Convergence rates of posterior distributions for non-i.i.d.\ observations. The Annals of Statistics, 35(1), 192--223

    2007

  17. [25]

    Ghoshal, A., & Honorio, J. (2017). Learning identifiable Gaussian Bayesian networks in polynomial time and sample complexity. In Advances in Neural Information Processing Systems 30, 6457--6466

    2017

  18. [26]

    E., Li, C., & Rabinovic, A

    Johnson, W. E., Li, C., & Rabinovic, A. (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics, 8(1), 118--127

    2007

  19. [27]

    Kalisch, M., & B \"u hlmann, P. (2007). Estimating high-dimensional directed acyclic graphs with the PC-algorithm. Journal of Machine Learning Research, 8, 613--636

    2007

  20. [28]

    Khare, K., Oh, S.-Y., & Rajaratnam, B. (2015). A convex pseudolikelihood framework for high dimensional partial correlation estimation with convergence guarantees. Journal of the Royal Statistical Society. Series B, 77(4), 803--825

    2015

  21. [29]

    Khare, K., Oh, S.-Y., Rahman, S., & Rajaratnam, B. (2019). A scalable sparse Cholesky based approach for learning high-dimensional covariance matrices in ordered data. Machine Learning, 108(12), 2061--2086

    2019

  22. [30]

    K., Slorach, E

    Kouros-Mehr, H., Bechis, S. K., Slorach, E. M., et al. (2008). GATA-3 links tumor differentiation and dissemination in a luminal breast cancer model. Cancer Cell, 13(2), 141--152

    2008

  23. [31]

    Kuipers, J., & Moffa, G. (2017). Partition MCMC for inference on acyclic digraphs. Journal of the American Statistical Association, 112(517), 282--299

    2017

  24. [32]

    Kuipers, J., Suter, P., & Moffa, G. (2022). Efficient sampling and structure learning of Bayesian networks. Journal of Computational and Graphical Statistics, 31(3), 639--650

    2022

  25. [33]

    W., Chen, Y., Shi, W., & Smyth, G

    Law, C. W., Chen, Y., Shi, W., & Smyth, G. K. (2014). voom: precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biology, 15(2), R29

    2014

  26. [34]

    T., Scharpf, R

    Leek, J. T., Scharpf, R. B., Bravo, H. C., et al. (2010). Tackling the widespread and critical impact of batch effects in high-throughput data. Nature Reviews Genetics, 11(10), 733--739

    2010

  27. [35]

    A., & Bhadra, A

    Li, Y., Craig, B. A., & Bhadra, A. (2019). The graphical horseshoe estimator for inverse covariance matrices. Journal of Computational and Graphical Statistics, 28(3), 747--757

    2019

  28. [36]

    Liu, H., Lafferty, J., & Wasserman, L. (2009). The nonparanormal: semiparametric estimation of high-dimensional undirected graphs. Journal of Machine Learning Research, 10, 2295--2328

    2009

  29. [37]

    Liu, C., & Martin, R. (2019). An empirical G -Wishart prior for sparse high-dimensional Gaussian graphical models. arXiv preprint arXiv:1912.03807

    2019 arXiv

  30. [38]

    B., Jordan, M

    Lopez, R., Regier, J., Cole, M. B., Jordan, M. I., & Yosef, N. (2018). Deep generative modeling for single-cell transcriptomics. Nature Methods, 15(12), 1053--1058

    2018

  31. [39]

    I., Huber, W., & Anders, S

    Love, M. I., Huber, W., & Anders, S. (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology, 15(12), 550

    2014

  32. [40]

    Madigan, D., York, J., & Allard, D. (1995). Bayesian graphical models for discrete data. International Statistical Review, 63(2), 215--232

    1995

  33. [41]

    Makalic, E., & Schmidt, D. F. (2016). A simple sampler for the horseshoe estimator. IEEE Signal Processing Letters, 23(1), 179--182

    2016

  34. [42]

    C., K \"u ffner, R., et al

    Marbach, D., Costello, J. C., K \"u ffner, R., et al. (2012). Wisdom of crowds for robust gene network inference. Nature Methods, 9(8), 796--804

    2012

  35. [43]

    A., Nemenman, I., Basso, K., et al

    Margolin, A. A., Nemenman, I., Basso, K., et al. (2006). ARACNE: an algorithm for the reconstruction of gene regulatory networks in a mammalian cellular context. BMC Bioinformatics, 7(Suppl 1), S7

    2006

  36. [44]

    A., Martin, R

    Maronna, R. A., Martin, R. D., Yohai, V. J., & Salibi \'a n-Barrera, M. (2019). Robust Statistics: Theory and Methods (with R), 2nd edition. Chichester: Wiley

    2019

  37. [45]

    Meinshausen, N., & B \"u hlmann, P. (2006). High-dimensional graphs and variable selection with the Lasso. The Annals of Statistics, 34(3), 1436--1462

    2006

  38. [46]

    O., & Coad, A

    Moneta, A., Entner, D., Hoyer, P. O., & Coad, A. (2013). Causal inference by independent component analysis: theory and applications. Oxford Bulletin of Economics and Statistics, 75(5), 705--730

    2013

  39. [47]

    P., & Lauritzen, S

    Mortera, J., Dawid, A. P., & Lauritzen, S. L. (2003). Probabilistic expert systems for DNA mixture profiling. Theoretical Population Biology, 63(3), 191--205

    2003

  40. [48]

    Peters, J., & B \"u hlmann, P. (2014). Identifiability of Gaussian structural equation models with equal error variances. Biometrika, 101(1), 219--228

    2014

  41. [49]

    G., & Scott, J

    Polson, N. G., & Scott, J. G. (2010). Shrink globally, act locally: sparse Bayesian regularization and prediction. In Bayesian Statistics 9, J. M. Bernardo et al.\ (eds.), pp. 501--538. Oxford University Press

    2010

  42. [50]

    D., & Oshlack, A

    Robinson, M. D., & Oshlack, A. (2010). A scaling normalization method for differential expression analysis of RNA-seq data. Genome Biology, 11(3), R25

    2010

  43. [51]

    Shojaie, A., & Michailidis, G. (2010). Penalized likelihood methods for estimation of sparse high-dimensional directed acyclic graphs. Biometrika, 97(3), 519--538

    2010

  44. [52]

    Spirtes, P., Glymour, C., & Scheines, R. (2000). Causation, Prediction, and Search, 2nd edition. Cambridge, MA: MIT Press

    2000

  45. [53]

    The Cancer Genome Atlas Network. (2012). Comprehensive molecular portraits of human breast tumours. Nature, 490(7418), 61--70

    2012

  46. [54]

    L., Kleijn, B

    van der Pas, S. L., Kleijn, B. J. K., & van der Vaart, A. W. (2014). The horseshoe estimator: posterior concentration around nearly black vectors. Electronic Journal of Statistics, 8(2), 2585--2618

    2014

  47. [55]

    Vershynin, R. (2018). High-Dimensional Probability: An Introduction with Applications in Data Science. Cambridge: Cambridge University Press

    2018

  48. [56]

    Wainwright, M. J. (2009). Information-theoretic limits on sparsity recovery in the high-dimensional and noisy setting. IEEE Transactions on Information Theory, 55(12), 5728--5741

    2009

  49. [57]

    Wang, H. (2012). Bayesian graphical Lasso models and efficient posterior computation. Bayesian Analysis, 7(4), 867--886

    2012

  50. [58]

    Warnick, R., Guindani, M., Erhardt, E., Allen, E., Calhoun, V., & Vannucci, M. (2018). A Bayesian approach for estimating dynamic functional network connectivity in fMRI data. Journal of the American Statistical Association, 113(521), 134--151

    2018

  51. [59]

    Welling, M., & Teh, Y. W. (2011). Bayesian learning via stochastic gradient Langevin dynamics. In Proceedings of the 28th International Conference on Machine Learning (ICML), pp. 681--688

    2011

  52. [60]

    West, M. (1987). On scale mixtures of normal distributions. Biometrika, 74(3), 646--648

    1987

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.