Pith. sign in

REVIEW 4 major objections 4 minor 50 references

This paper establishes the first consistency guarantee for multivariate random forests, proving that permutation-based variable importance retains every truly influential feature with probability tending to one as sample size grows.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 06:25 UTC pith:JSVUFNKJ

load-bearing objection Real conditional screening theorem for MRF-PVIM, but the algorithm as written is undefined (singular node covariances) and the proved threshold rule isn't the one run; worth refereeing, needs major revision. the 4 major comments →

arxiv 2607.21880 v1 pith:JSVUFNKJ submitted 2026-07-24 stat.ME math.STstat.TH

Theoretical Properties of Multivariate Random Forest in Feature Selection and its Application to Facial Morphology-Gene Detection

Yangsheng Wang , Samruddhi Thakar , Anton Schick , Guifang Fu This is my paper
classification stat.ME math.STstat.TH MSC 62G2062H3062H12
keywords multivariate random forestspermutation variable importancesure screeningfeature selectionincomplete U-statisticsGWASfacial morphologyhigh-dimensional inference
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tries to put multivariate random forests (MRF) on firm theoretical ground for feature selection. Its central claim is that the permutation-based variable importance measure (PVIM) has the sure screening property: when the sample size grows, with probability approaching one, the procedure retains every feature that truly influences a multivariate outcome. The proof models PVIM as an incomplete U-statistic that averages over three layers of randomness—subsampled subjects, subsampled features at each split, and random permutations of out-of-bag observations. If the claim holds, MRF becomes a joint, interaction-aware alternative to independence screening, and the paper demonstrates its utility on a genome-wide scan of human facial morphology.

Core claim

The paper establishes the first consistency guarantee for permutation-based variable importance in multivariate random forests. Stated on the paper's own terms: under moment conditions C1–C3 on tree prediction errors, the estimator's maximum variance over all p features is O(R_n), and if Card(D)R_n/c0^2 → 0, then P(D ⊆ D̂n) → 1, where D is the set of truly influential features and D̂n the selected set. The selection rule keeps features whose estimated PVIM exceeds half of a population threshold c0 separating signal from noise. The proof decomposes the PVIM estimator into a centered randomness term, a bias term, and an incomplete U-statistic term, bounding each via combinatorial counting of o

What carries the argument

The central object is the permutation-based variable importance measure (PVIM) of multivariate random forests, defined as the change in out-of-bag prediction error after permuting a single feature, averaged over the ensemble. The proof reshapes PVIM as an incomplete U-statistic averaging over three independent layers of randomness—subsampling of subjects, subsampling of candidate features at each split, and permutation of features on out-of-bag samples—and bounds its variance by decomposing the estimator into a centered per-tree term, a bias term, and a U-statistic term (A_nj, B_nj, U_nj). The splitting rule itself uses the within-node sum of squared Mahalanobis distances, with the inverse o

Load-bearing premise

The result holds only if conditions C1 and C2—unverified moment bounds on the prediction errors of the individual trees—are satisfied, and the splitting rule as written inverts a within-node covariance matrix that is singular whenever a node has fewer observations than outcome variables, a situation the paper's own stopping rule (minimum five observations) can create when q is ten or fifty.

What would settle it

Run the algorithm exactly as defined on the paper's Simulation 1 (n=200, q=10 outcomes, stopping at five observations per child): any node with fewer than ten observations yields a singular V(A), so the Mahalanobis split criterion cannot be computed without an unspecified pseudo-inverse or ridge; demonstrating this on a single replicate is enough to show the procedure is not self-contained. Alternatively, estimate the left-hand sides of conditions C1 and C2 from simulated trees and check whether the variance actually scales as O(n) and O(k_n^α); if not, the theorem's rate (6) does not follow.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • With the sure screening property, MRF-based feature selection can serve as a screening step that, as n grows, does not drop any truly influential feature.
  • The joint, permutation-based importance accounts for multicollinearity and interactions in a way independence screening cannot; the facial-morphology application illustrates this by identifying hub genes such as GPD1L in males and GNG4 in females.
  • The max-ratio thresholding rule, combined with the theorem, yields parsimonious selections: in simulations MRF recovers all true features with a mean minimum selection size of about 7 out of 500 features.
  • The two-step LASSO-then-MRF strategy is validated in ultrahigh-dimensional settings, achieving a 90% overall success rate at p=100,000 while keeping false selection rates low.
  • The theoretical result applies even when the number of features, the set of true features, and the threshold c0 grow with n, covering the high-dimensional regime beyond classical fixed-parameter screening.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The unverified moment conditions C1–C2 are the bridge between the algorithm and the guarantee; one could test them empirically, and if they fail for the Mahalanobis splitting rule, the sure-screening conclusion would not apply to the implemented method.
  • The two-step LASSO-then-MRF strategy is itself a hybrid that trades joint screening for computational feasibility; a natural extension would be to derive a screening guarantee for the two-step selector rather than only the MRF stage.
  • The 'hub' interactions (GPD1L in males, GNG4 in females) are the kind of hypothesis a joint method can generate that single-SNP tests cannot; one could validate them with gene-expression or knockdown experiments.
  • If PVIM variance can be estimated, the max-ratio threshold could be replaced by an inference-based cutoff, connecting this screening result to formal statistical tests for feature importance.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes multivariate random forest (MRF) with permutation-based variable importance (PVIM) as a joint feature-screening tool for multivariate outcomes. It states a theorem (Section 2.2) claiming max_j Var(hat-Lambda_n^{(j)}) = O(R_n) with R_n = max{k_n^{2+alpha}/n, n/m_n}, and a sure-screening conclusion P(D subset of hat-D_n) -> 1 under Condition 1, Conditions C1-C3, and Card(D) R_n / c0^2 -> 0. The proof uses an incomplete U-statistic decomposition to aggregate subsampling of subjects, mtry feature subsampling, and permutation randomness. The paper also reports simulations comparing MRF with CCA, LASSO, and five SIS methods, and applies MRF (after LASSO pre-screening) to a facial-morphology GWAS with 2,342 subjects and 453,273 SNPs. The central claim is the 'first consistency guaranty for MRF'.

Significance. If the theorem were established for a well-defined algorithm, the incomplete-U-statistic variance decomposition and the screening result would fill a recognized gap in the multivariate random forest literature. The paper is also honest about practical limitations: it explicitly states that direct MRF 'breaks down' at ultrahigh dimensionality and that a LASSO pre-screening step is needed. The heterogeneous simulation designs are a constructive contribution. However, as it stands, the theorem is conditional on unverified high-level moment conditions, the algorithm as defined cannot be executed in the paper's own simulation settings because the splitting criterion uses an inverse of a singular covariance matrix, and the practical selection rule differs from the theoretical threshold rule. These issues are load-bearing for the paper's central claim, so the significance cannot be credited in its present form.

major comments (4)
  1. [Section 2, definition of SSE(A)] The splitting rule uses SSE(A) = sum_i (Y_i - bar-Y(A))^T V(A)^{-1} (Y_i - bar-Y(A)), where V(A) is 'the covariance matrix of the outcome vector estimated using the observations in node A.' The stopping rule only forbids child nodes with fewer than five observations. In Simulation Study 1 and the GWAS, q = 10; in Simulation Study 2, q = 5. Therefore nodes of size between 5 and q are legal, and for such nodes V(A) has rank at most n_A - 1 < q, so V(A) is singular and V(A)^{-1} does not exist. For example, a parent node of size 12 split into children of size 5 and 7 must evaluate SSE(A_L) and SSE(A_R) on singular covariance matrices. No regularization, pseudo-inverse, shrinkage, or minimum-node-size greater than q is stated anywhere. Thus the algorithm as described cannot be run in the paper's own simulations, and the theorem has no well-defined object. This is not a minor implementation d
  2. [Theorem, Section 2.2, Conditions C1-C2] Conditions C1 and C2 are high-level moment bounds on the tree prediction errors g_n and g_{nj}. The paper never verifies these conditions for the proposed MRF splitting rule under any concrete data-generating process. Consequently the theorem states: if the variance of the PVIM estimator is O(R_n) and true PVIMs are separated from zero by c0, then thresholding at c0/2 works. The substantive part, namely proving the variance bound for this specific algorithm, is assumed rather than established. The abstract's phrase 'under mild regularity conditions' overstates the status of C1-C2; as written, these are strong, unverified assumptions on an estimator whose construction is itself not fully specified.
  3. [Eq. (5) vs. Sections 3.1 and 4] The theoretical selection rule is hat-D_n = {j : hat-Lambda_n^{(j)} > c0/2}, where c0 is the constant in Condition 1. The empirical rule used in all simulations and in the real data is the max-ratio rule: retain features that fall in the top 5% by PVIM and simultaneously among the top five max-ratio thresholds. These are different rules, and c0 is unknown in practice. The theorem provides no guarantee for the max-ratio rule. Thus the reported individual/overall success rates and the GWAS hits are produced by a procedure for which the consistency theorem is not proved.
  4. [Sections 3.3.3 and 4, LASSO pre-screening] The real-data pipeline and Simulation Study 3 use a LASSO pre-screening step before MRF; Section 3.3.3 states that direct MRF 'breaks down' at p = 100,000, and Section 4 reduces 453,273 SNPs to about 66,979 (male) and 97,786 (female) by LASSO before MRF is applied. The theorem is stated and proved only for MRF without pre-screening. No theoretical result covers the LASSO+MRF pipeline, so the GWAS findings are not backed by the paper's consistency guarantee. This gap is acknowledged in the practical discussion but not addressed in the theory.
minor comments (4)
  1. [Appendix A, proof of part (2)] The statement that 'b1 contains at most k_n^2 m_n^2 non-zero elements' appears to undercount; a more conservative count gives m_n^2 n k_n non-zero terms. Under C3 this still yields the stated O(R_n) rate, but the current line should be corrected.
  2. [Section 3.1, max-ratio rule] The 'top five max-ratio thresholds' rule is defined only by reference to Zhao and Fu [48]. A self-contained definition of the ratio and of how the top-five rule interacts with the 5% cutoff would make the empirical procedure reproducible without consulting another paper.
  3. [Throughout] 'Cauchy-Schwartz' should be 'Cauchy-Schwarz' in Appendix A.
  4. [Tables 3, 5, and 7] The tables reporting individual and overall success rates are visually cramped and hard to read; separating the success rates for each true feature into a clearer layout would improve presentation.

Circularity Check

0 steps flagged

No material circularity; the screening guarantee is a conditional threshold consequence of a standard concentration bound, not a self-referential derivation.

full rationale

The paper's derivation chain is not circular in the sense of equation X being defined from Y and then Y being 'predicted' from X. The screening property (8) follows by Chebyshev from the variance bound (6), and (6) follows from standard incomplete U-statistic variance calculations under the stated moment conditions C1–C2. Condition 1 is an oracle separability assumption ('there exists a constant c0>0 such that for every j in D, Λ(j)>c0') parallel to minimum-marginal-utility conditions in the SIS literature; it is assumed, not fitted, and the theorem's conclusion is conditional on it. The practical threshold in the simulations and real data is the max-ratio rule cited from Zhao and Fu [48], a self-citation (Fu is an author), but that rule is not used in the theorem and is an externally published procedure, so it is not load-bearing for the consistency claim. The most serious issue in the manuscript is not circularity: the splitting rule SSE(A)=Σ(Y_i−Ȳ(A))ᵀV^{-1}(A)(Y_i−Ȳ(A)) requires the inverse of the within-node outcome covariance matrix, while the stated stopping rule permits legal nodes with 5≤n_A≤q, making V(A) singular in the q=10 and q=5 simulations and in the GWAS; moreover C1–C2 are never verified for the proposed splitting rule. These are correctness/completeness gaps, not a derivation that reduces to its own inputs. The low score reflects the conditional/oracle structure of the theorem and the minor non-load-bearing self-citation, not a circular construction.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 0 invented entities

The central claim rests on a lower-bound signal assumption, unverified moment conditions on the tree predictor, and an invertibility assumption on node-level covariance matrices; no new entities are introduced.

free parameters (6)
  • c0 (lower bound on true PVIM) = unknown
    Condition 1 and selection rule (5) require c0>0 with Λ(j)>c0 for all true j; c0 is not estimated.
  • kn (subsample size per tree) = not specified
    Theorem requires k_n^{2+α}/n→0; no constructive choice is given for finite n.
  • mn (number of trees) = not specified
    Theorem requires n/m_n→0; no guidance for choosing m_n in practice.
  • mtry (candidate features per split) = not specified
    Algorithm parameter; values used in simulations and real data are not reported.
  • max-ratio threshold rule (top 5%, top-five max-ratio) = data-dependent
    Used in simulations and GWAS instead of c0/2; no theory shows it inherits the screening property.
  • LASSO pre-screening penalty = not specified
    Reduces 453,273 SNPs to 66,979/97,786; λ selection is not reported.
axioms (4)
  • domain assumption Within-node outcome covariance V(A) is invertible (or has an unstated regularization) for every node
    SSE(A) is defined with V^{-1}(A); terminal nodes have as few as five observations while q=10 or 50, so V(A) is singular in the paper's own settings.
  • domain assumption Conditions C1-C2: E[(g_n-ḡ_n)^2]=O(n) and E[ḡ_n^2]=O(k_n^α), uniformly in j
    These moment bounds on MRF predictions are assumed in the theorem and never derived from the splitting rule.
  • domain assumption Condition 1: all true features have population PVIM above c0
    The screening guarantee only covers features separated from zero by a known threshold; no diagnostic for this is provided.
  • standard math Incomplete U-statistic variance formula (Eq. 12) applies to the with-replacement subsampling scheme
    Proof of step (3) relies on the quoted variance formula; the paper's subsampling is with replacement, making the equality questionable.

pith-pipeline@v1.3.0-alltime-deepseek · 21951 in / 19186 out tokens · 174676 ms · 2026-08-01T06:25:11.108912+00:00 · methodology

0 comments
read the original abstract

This work establishes a theoretical foundation for joint feature selection with multivariate outcomes, positioning the permutation-based variable importance measure (PVIM) of multivariate random forests (MRF) as a principled tool for high-dimensional feature selection. We establish the first consistency guaranty for MRF, showing that it retains all truly influential features with probability tending to one as the sample size grows to infinity under mild regularity conditions. Incomplete U-statistics is employed to incorporate three layers of randomness: subsampling of subjects for training each tree, subsampling of features at each split, and permutation of each feature for the out-of-bag (OOB) samples. Unlike independence-based screening that evaluates each feature in isolation, PVIM is a joint screening approach that accounts for multicollinearity, nonlinear, high-order interactions, and subject heterogeneity via ensemble aggregation. Moreover, we demonstrate the practical utility of MRF through a genome-wide association study (GWAS) of human facial morphology (with 2,342 subjects and 453,273 SNPs), where MRF identifies several novel loci and interaction hubs that extend prior findings. Extensive simulations show that MRF accurately identifies truly influential signals while producing parsimonious feature sets with well controlled false selection rates, outperforming canonical correlation analysis (CCA) and several other independence multivariate screening approaches. In addition, we also propose a novel simulation framework, including image outcomes, that more closely mimic the intricate nature of real-world data and provide rigorous testbeds for machine learning research.

Figures

Figures reproduced from arXiv: 2607.21880 by Anton Schick, Guifang Fu, Samruddhi Thakar, Yangsheng Wang.

Figure 1
Figure 1. Figure 1: Illustration of node-wise feature subsampling in the [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Four examples of synthetic human face simulations. [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Boxplots of the minimum selection size S in Simulation Study 1 across 100 replicated simulation datasets. The eight colors represent the eight approaches [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Boxplots of the false selection rate T for Simulation Study 1 across 100 replicated datasets. Left panel: Results for all eight approaches (each color corresponds to one method). Right panel: Zoomed-in results for the seven approaches excluding LASSO. tables and figures for this additional simulation are omitted, as they exhibit patterns similar to those already presented [PITH_FULL_IMAGE:figures/full_fig… view at source ↗
Figure 5
Figure 5. Figure 5: Boxplots of the minimum selection size S in Simulation Study 2 across 100 replicated simulation datasets. The eight colors represent the eight approaches [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Boxplots of the false selection rate T for Simulation Study 2 across 100 replicated datasets. Left panel: Results for all eight approaches (each color corresponds to one method). Right panel: Zoomed-in results for the seven approaches excluding LASSO. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Boxplots of the minimum selection size S in Simulation Study 3 across 10 replicated simulation datasets. Left panel: Results for the two scenarios (each color corresponds to one method). Right panel: Zoomed-in results for the two scenarios, excluding the most extreme outlier value of CCA [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Boxplots of the false selection rate T for Simulation Study 3 across 10 replicated simulation datasets. additional evidence on false selection rates: LASSO+MRF exhibits the narrowest interquartile range, underscoring its efficiency and accuracy in avoiding false selections. In summary, these experiments underscore the practical advantages of incorporating LASSO as a pre-screening step in ultrahigh-dimensio… view at source ↗
Figure 9
Figure 9. Figure 9: Permutation-based variable importance measures for all features in the male group. [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Permutation-based variable importance scores for all features in the female group. [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 15 canonical work pages · 1 internal anchor

  1. [1]

    Albers, H

    M. Albers, H. Kranz, I. Kober, et al. Automated yeast two-hybrid screening for nuclear receptor-interacting proteins.Molecular & Cellular Proteomics, 4(2):205–213, 2005

    2005

  2. [2]

    Analysis and comparison of feature selection methods towards performance and stability.Expert Systems with Applications, 249:123667, 2024

    Matheus Cezimbra Barbieri, Bruno Iochins Grisci, and Márcio Dorn. Analysis and comparison of feature selection methods towards performance and stability.Expert Systems with Applications, 249:123667, 2024. ISSN 0957-4174. doi:https://doi.org/10.1016/j.eswa.2024.123667. URL https://www.sciencedirect.com/science/article/pii/S0957417424005335

    arXiv 2024

  3. [3]

    G. Biau. Analysis of a random forests model.Journal of Machine Learning Research, 13: 1063–1095, 2012

    2012

  4. [4]

    Blum and Pat Langley

    Avrim L. Blum and Pat Langley. Selection of relevant features and examples in machine learning.Artificial Intelligence, 97(1):245–271, 1997. ISSN 0004-3702. doi:https://doi.org/10.1016/S0004-3702(97)00063-5. URLhttps://www.sciencedirect.com/ science/article/pii/S0004370297000635. Relevance

    work page doi:10.1016/s0004-3702(97)00063-5 1997

  5. [5]

    A randomized subspace-based approach for dimensionality reduction and important variable selection.Jour- nal of Machine Learning Research, 24(76):1–31, 2023

    Di Bo, Hoon Hwangbo, Vinit Sharma, Corey Arndt, and Stephanie TerMaath. A randomized subspace-based approach for dimensionality reduction and important variable selection.Jour- nal of Machine Learning Research, 24(76):1–31, 2023. URLhttp://jmlr.org/papers/v24/ 21-1046.html

    2023

  6. [6]

    Cajo J. F. Ter Braak. Canonical correspondence analysis: A new eigenvector technique for multivariate direct gradient analysis.Ecology, 67(5):1167–1179, 1986. ISSN 00129658, 19399170. URLhttp://www.jstor.org/stable/1938672

    arXiv 1986

  7. [7]

    Springer, 2011

    Peter Bühlmann and Sara Van De Geer.Statistics for High-Dimensional Data: Methods, Theory and Applications. Springer, 2011

    2011

  8. [8]

    A survey on feature selection meth- ods.Computers & Electrical Engineering, 40(1):16–28, 2014

    Girish Chandrashekar and Ferat Sahin. A survey on feature selection meth- ods.Computers & Electrical Engineering, 40(1):16–28, 2014. ISSN 0045-7906. doi:https://doi.org/10.1016/j.compeleceng.2013.11.024. URL https://www.sciencedirect. com/science/article/pii/S0045790613003066. 40th-year commemorative issue

    doi:10.1016/j.compeleceng.2013.11.024 2014

  9. [9]

    Improved facial outcome assessment using a 3d anthropometric mask.International journal of oral and maxillofacial surgery, 41:324–30, 11

    Peter Claes, Mark Walters, and John Clement. Improved facial outcome assessment using a 3d anthropometric mask.International journal of oral and maxillofacial surgery, 41:324–30, 11

  10. [10]

    Liberton, Katleen Daniels, Kerri Matthes Rosana, Ellen E

    Peter Claes, Denise K. Liberton, Katleen Daniels, Kerri Matthes Rosana, Ellen E. Quillen, LaurelN.Pearson, Brian McEvoy, MarcBauchet, Arslan A.Zaidi, WeiYao, HuaTang, GregoryS. Barsh, Devin M. Absher, David A. Puts, Jorge Rocha, Sandra Beleza, Rinaldo W. Pereira, Gareth Baynam, Paul Suetens, Dirk Vandermeulen, Jennifer K. Wagner, James S. Boster, and Mark...

    work page doi:10.1371/journal.pgen.1004224 2014

  11. [11]

    White, Tomek Swigut, Dzemila Sero, Jiarui Li, Myoung Keun Lee, Arslan Zaidi, Brooke C

    Peter Claes, Jasmien Roosenboom, Julie D. White, Tomek Swigut, Dzemila Sero, Jiarui Li, Myoung Keun Lee, Arslan Zaidi, Brooke C. Mattern, Corey Liebowitz, Laurel Pearson, Tomás González, Elizabeth J. Leslie, Jenna C. Carlson, Ekaterina Orlova, Paul Suetens, Dirk Vandermeulen, Eleanor Feingold, Mary L. Marazita, John R. Shaffer, Joanna Wysocka, Mark D. Shr...

    2018

  12. [12]

    Conte, A

    I. Conte, A. Capasso, M. Avellino, A. M. Ciacci, V. De Simone, L. Nitsch, M. M. Miano, and S. D. Acierno. Copy number variations in SATB2 associated with orofacial clefts.American Journal of Medical Genetics Part A, 170(4):896–903, 2016. doi:10.1002/ajmg.a.37484

    work page doi:10.1002/ajmg.a.37484 2016

  13. [13]

    Minimum redundancy feature selection from microarray gene expression data

    Chris Ding and Hanchuan Peng. Minimum redundancy feature selection from microarray gene expression data. InProceedings of the 2nd IEEE Computer Society Bioinformatics Conference (CSB), pages 523–528. IEEE, 2003. doi:10.1109/CSB.2003.1227396

    Pith/arXiv arXiv 2003

  14. [14]

    Sure independence screening for ultrahigh dimensional fea- ture space.Journal of the Royal Statistical Society: Series B (Statistical Methodology), 70(5):849–911, 2008

    Jianqing Fan and Jinchi Lv. Sure independence screening for ultrahigh dimensional fea- ture space.Journal of the Royal Statistical Society: Series B (Statistical Methodology), 70(5):849–911, 2008. doi:https://doi.org/10.1111/j.1467-9868.2008.00674.x. URL https: //rss.onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-9868.2008.00674.x

    arXiv 2008

  15. [15]

    Manuel A. R. Ferreira and Shaun M. Purcell. A multivariate test of association.Bioinformatics, 25(1):132–133, January 2009. doi:10.1093/bioinformatics/btn563

    work page doi:10.1093/bioinformatics/btn563 2009

  16. [16]

    C. D. Go, J. D. R. Knight, A. Rajasekharan, et al. A proximity-dependent biotinylation map of a human cell.Nature, 595:120–124, 2021

    2021

  17. [17]

    X. Guo, R. Li, W. Liu, et al. Stable correlation and robust feature screening.Science China Mathematics, 65(1):153–168, 2022

    2022

  18. [18]

    An introduction to variable and feature selection.J

    Isabelle Guyon and André Elisseeff. An introduction to variable and feature selection.J. Mach. Learn. Res., 3(null):1157–1182, March 2003. ISSN 1532-4435

    2003

  19. [19]

    Zadeh.Feature Extraction: Foundations and Applications (Studies in Fuzziness and Soft Computing)

    Isabelle Guyon, Steve Gunn, Masoud Nikravesh, and Lotfi A. Zadeh.Feature Extraction: Foundations and Applications (Studies in Fuzziness and Soft Computing). Springer-Verlag, Berlin, Heidelberg, 2006. ISBN 3540354875

    2006

  20. [20]

    Fengler.Canonical Correlation Analysis, pages 443–454

    Wolfgang Karl Härdle, Léopold Simar, and Matthias R. Fengler.Canonical Correlation Analysis, pages 443–454. Springer International Publishing, Cham, 2024. ISBN 978-3-031-63833-6. doi:10.1007/978-3-031-63833-6_16. URL https://doi.org/10.1007/978-3-031-63833-6_16

    work page doi:10.1007/978-3-031-63833-6_16 2024

  21. [21]

    Hardoon, Sandor R

    David R. Hardoon, Sandor R. Szedmak, and John R. Shawe-taylor. Canonical correlation analysis: An overview with application to learning methods.Neural Comput., 16(12):2639–2664, December 2004. ISSN 0899-7667. doi:10.1162/0899766042321814. URLhttps://doi.org/10. 1162/0899766042321814

    doi:10.1162/0899766042321814 2004

  22. [22]

    Johnson and Dean W

    Richard A. Johnson and Dean W. Wichern.Applied Multivariate Statistical Analysis. Pearson Prentice Hall, Upper Saddle River, NJ, 6th edition, 2007

    2007

  23. [23]

    Ron Kohavi and George H. John. Wrappers for feature subset selection.Artificial Intelligence, 97 (1):273–324, 1997. ISSN 0004-3702. doi:https://doi.org/10.1016/S0004-3702(97)00043-X. URL https://www.sciencedirect.com/science/article/pii/S000437029700043X. Relevance

    doi:10.1016/s0004-3702(97)00043-x 1997

  24. [24]

    Shap-select: Lightweight feature selection using shap values and regression, 2024

    Egor Kraev, Baran Koseoglu, Luca Traverso, and Mohammed Topiwalla. Shap-select: Lightweight feature selection using shap values and regression, 2024. URLhttps://arxiv.org/ abs/2410.06815. 29

    Pith/arXiv arXiv 2024

  25. [25]

    Trevino, Jiliang Tang, and Huan Liu

    Jundong Li, Kewei Cheng, Suhang Wang, Fred Morstatter, Robert P. Trevino, Jiliang Tang, and Huan Liu. Feature selection: A data perspective.ACM Computing Surveys, 50(6):1–45, December 2017. ISSN 0360-0300. doi:10.1145/3136625

    doi:10.1145/3136625 2017

  26. [26]

    R. Li, W. Zhong, and L. Zhu. Feature screening via distance correlation learning.Journal of the American Statistical Association, 107(499):1129–1139, 2012

    2012

  27. [27]

    W. Liu, Y. Ke, J. Liu, and R. Li. Model-free feature screening and fdr control with knockoff features.Journal of the American Statistical Association, 117(537):428–443, 2020

    2020

  28. [28]

    A. T. Marees, H. de Kluiver, S. Stringer, F. Vorspan, E. Curis, C. Marie-Claire, and E. M. Derks. A tutorial on conducting genome-wide association studies: Quality control and statis- tical analysis.International Journal of Methods in Psychiatric Research, 27(2):e1608, 2018. doi:10.1002/mpr.1608. URLhttps://doi.org/10.1002/mpr.1608

    work page doi:10.1002/mpr.1608 2018

  29. [29]

    Mentch and G

    L. Mentch and G. Hooker. Quantifying uncertainty in random forests via confidence intervals and hypothesis tests.Journal of Machine Learning Research, 17(1):841–881, 2016

    2016

  30. [30]

    A. J. Osborne, A. Bierzynska, E. Colby, and et al. Multivariate canonical correlation analysis identifies additional genetic variants for chronic kidney disease.npj Systems Biology and Applications, 10(28):28, 2024. doi:10.1038/s41540-024-00350-8

    work page doi:10.1038/s41540-024-00350-8 2024

  31. [31]

    W. Pan, X. Wang, W. Xiao, and H. Zhu. A generic sure independence screening procedure. Journal of the American Statistical Association, 114(526):928–937, 2019

    2019

  32. [32]

    J. K. Pickrell et al. Detection and interpretation of shared genetic influences on 42 human traits.Nature Genetics, 48:709–717, 2016

    2016

  33. [33]

    Kempa-Liehr, and Justin M

    Nanang Pudjihartono, Tejaswini Fadason, Andreas W. Kempa-Liehr, and Justin M. O’Sullivan. Areviewoffeatureselectionmethodsformachinelearning-baseddiseaseriskprediction.Frontiers in Bioinformatics, 2:927312, June 2022. doi:10.3389/fbinf.2022.927312

    arXiv 2022

  34. [34]

    Exploring elastic net and multivariate regression

    Matthias Raess. Exploring elastic net and multivariate regression.arXiv preprint arXiv:1607.06763, July 2016. doi:10.48550/arXiv.1607.06763. URLhttps://arxiv.org/abs/ 1607.06763

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1607.06763 2016

  35. [35]

    Rainger, H

    J. Rainger, H. Bengani, T. Campbell, M. Bergbaum, J. J. Sankaralingam, E. A. Walker, T. Koko- tas, S. Alkhunaizi, L. S. Lampe, R. Redmond, T. Prescott, C. A. Anderson, R. D. Campbell, C. P. Ponting, and D. FitzPatrick. Disruption of SATB2 causes micrognathia and clefting in humans and mice.Human Molecular Genetics, 23(10):2564–2577, 2014. doi:10.1093/hmg/ddt644

    work page doi:10.1093/hmg/ddt644 2014

  36. [36]

    Rencher and William F

    Alvin C. Rencher and William F. Christensen.Methods of Multivariate Analysis. John Wiley & Sons, Hoboken, NJ, 3rd edition, 2012

    2012

  37. [37]

    Overfitting in making comparisons between variable selection methods.Journal of Machine Learning Research, 3:1371–1382, 2003

    Juha Reunanen. Overfitting in making comparisons between variable selection methods.Journal of Machine Learning Research, 3:1371–1382, 2003. URLhttp://www.jmlr.org/papers/v3/ reunanen03a.html

    2003

  38. [38]

    A review of feature selection tech- niques in bioinformatics.Bioinformatics, 23(19):2507–2517, August 2007

    Yvan Saeys, Iñaki Inza, and Pedro Larrañaga. A review of feature selection tech- niques in bioinformatics.Bioinformatics, 23(19):2507–2517, August 2007. ISSN 1367-

    2007

  39. [39]

    Scornet, G

    E. Scornet, G. Biau, and J.-P. Vert. Consistency of random forests.The Annals of Statistics, 43(4):1716–1741, 2015

    2015

  40. [40]

    Seoane, Colin Campbell, Ian N

    Jose A. Seoane, Colin Campbell, Ian N. Day, Juan P. Casas, and Tom R. Gaunt. Canonical correlation analysis for gene-based pleiotropy discovery.PLoS Computational Biology, 10(10): e1003876, October 2014. doi:10.1371/journal.pcbi.1003876

    work page doi:10.1371/journal.pcbi.1003876 2014

  41. [41]

    Sheehan-Rooney, B

    K. Sheehan-Rooney, B. Swartz, J. Zhao, J. B. McGonnell, J. C. Liu, and T. F. Richman. Satb2 is expressed in the medial nasal, maxillary and mandibular processes during facial development in mouse, chick, and zebrafish.Developmental Dynamics, 239(7):1789–1800, 2010. doi:10.1002/dvdy.22309

    work page doi:10.1002/dvdy.22309 2010

  42. [42]

    Variable importance measures for multivariate random forests.Journal of Data Science, 23(1):243–263, 2025

    Sharmistha Sikdar, Giles Hooker, and Vrinda Kadiyali. Variable importance measures for multivariate random forests.Journal of Data Science, 23(1):243–263, 2025. ISSN 1680-743X. doi:10.6339/24-JDS1152

    work page doi:10.6339/24-jds1152 2025

  43. [44]

    Synergy, redundancy, and multivariate information measures: An experimentalist’s perspective.Journal of computational neuroscience, 36, 07 2013

    Nicholas Timme, Wesley Alford, Benjamin Flecker, and John Beggs. Synergy, redundancy, and multivariate information measures: An experimentalist’s perspective.Journal of computational neuroscience, 36, 07 2013. doi:10.1007/s10827-013-0458-4

    work page doi:10.1007/s10827-013-0458-4 2013

  44. [45]

    Some properties of two measures of multivariate association.Psychometrika, 53(1):109–122, 1988

    Willem van den Burg and Charles Lewis. Some properties of two measures of multivariate association.Psychometrika, 53(1):109–122, 1988. doi:10.1007/BF02294197

    work page doi:10.1007/bf02294197 1988

  45. [46]

    A. W. van der Vaart.Asymptotic Statistics. Cambridge Books. Cambridge University Press, Cambridge, January 2000. ISBN 9780521784504

    2000

  46. [47]

    J. D. White, K. Indencleef, S. Naqvi, et al. Insights into the genetic architecture of the human face.Nature Genetics, 53(1):45–53, 2021

    2021

  47. [48]

    Zhao and G

    S. Zhao and G. Fu. Distribution-free and model-free multivariate feature screening via multi- variate rank distance correlation.Journal of Multivariate Analysis, 192:105081, 2022. 31

    2022

  48. [2011]

    doi:10.1016/j.ijom.2011.10.019

    work page doi:10.1016/j.ijom.2011.10.019 2011

  49. [2013]

    URLhttps://arxiv.org/abs/1311.6529

    Pith/arXiv arXiv

  50. [4803]

    URL https://doi.org/10.1093/bioinformatics/ btm344

    doi:10.1093/bioinformatics/btm344. URL https://doi.org/10.1093/bioinformatics/ btm344. 30

    doi:10.1093/bioinformatics/btm344