Pith. sign in

REVIEW 3 major objections 4 minor 130 references

DAIF: A Data-Driven Intermediate Fusion Framework for Multimodal Supervised Learning via Approximate Message Passing

T0 review · 3 major / 4 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 This paper proves that learning fusion granularity from data costs nothing asymptotically: clustering modalities with CKA and running clusterwise empirical-Bayes AMP recovers the same oracle-level latent representations and predictions.

desk verdict A substantive AMP-based fusion method with honest theory, but the data-adaptive fusion claim rests on a hard clique-separation assumption the real-data experiments only weakly support. read the letter →

arxiv 2608.02769 v1 pith:77MSUUYU submitted 2026-08-03 stat.ME cs.LGmath.STstat.MLstat.TH

Sagnik Nandy , Samriddha Lahiry , Pragya Sur , Subhabrata Sen This is my paper · ORCID
classification stat.MEcs.LGmath.STstat.MLstat.TH MSC 62H3062H2562F12
keywords dataadaptiveintermediatefusionmultimodalsupervisedlearningapproximatemessagepassingempiricalBayescenteredkernelalignmenthierarchicalclusteringstateevolutionfactormodel
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 tries to show that the right granularity for multimodal fusion—which modalities to integrate and which to keep separate—can be chosen from the data itself without sacrificing statistical optimality. It proposes DAIF, a pipeline that measures pairwise dependence between modalities with a kernel-based score, clusters the modalities by that score, estimates a prior over latent signals inside each cluster, and then denoises the features with approximate message passing. The payoff, if the theorems hold, is that early fusion and late fusion become special cases of one adaptive procedure rather than competing architectural choices. The method is aimed at settings like single-cell multi-omics, where RNA, chromatin, and protein data are unequally dependent and a fixed fusion choice helps some cell types while hurting others.

What carries the argument

The central object is the clusterwise product prior $\mu = \bigotimes_{k=1}^K \mu_k$ over the latent signatures $(U_1,\ldots,U_m,\tilde U_1,\ldots,\tilde U_{\tilde m})$, under the factor model $X_h = n^{-1/2} U_h D_h V_h^\top + Z_h$ for high-dimensional modalities and $\tilde X_\ell = \tilde U_\ell L_\ell^\top + \tilde Z_\ell$ for low-dimensional ones. The mechanism that carries the argument has four moving parts: centered kernel alignment (CKA), a normalized kernel dependence measure, converted into a dissimilarity matrix for average-linkage hierarchical clustering; a clustered empirical-Bayes maximum likelihood estimator that estimates each $\mu_k$ from the leading singular vectors inside the estimated cluster; AMP recursions that alternate Onsager-corrected power iteration with posterior-mean denoising under the estimated clusterwise prior; and state-evolution recursions that track the covariance of the iterates. The key identity driving the theory is that after the Onsager correction the iterates behave like the true signals plus Gaussian noise with known covariance, so posterior-mean shrinkage is the Bayes-optimal denoiser, and the estimated-prior version converges to the oracle version by the uniform Lipschitz regularity of the denoisers.

What would settle it

Generate data with three modalities where two are strongly dependent and the third has small but nonzero population dependence with one of them, so the separation gap is absent, increase the sample size, and check whether hierarchical clustering still recovers the true group structure and whether the DAIF state evolution still matches an oracle that knows the correct non-clique dependence; if the oracle and DAIF divergences persist at large n, the separation condition is not merely technical.

Watch

Extended reading notes

Core claim

The central claim is that the granularity of multimodal fusion can be learned from data and that learning it costs nothing in the large-sample limit. Under a Bayesian factor model in which the cross-modal dependence of the latent subject signatures is exactly a partition into independent cliques, DAIF (1) estimates pairwise modality dependence with centered kernel alignment, (2) recovers the cliques by average-linkage hierarchical clustering, (3) estimates each cluster's prior by empirical Bayes, and (4) runs an approximate message passing (AMP) denoiser with clusterwise priors. The main theorems state that the estimated prior converges weakly to the true prior (Theorem 6.1), that the AMP iterates have the same asymptotic state evolution as oracle iterates built from the true priors (Theorem 6.2), and that the limiting reconstruction error equals the Bayes-optimal error, i.e., the error of the posterior expectation of the signal given all modalities (Theorem 6.3). A fourth theorem (Theorem 6.4) gives the error-in-variables representation of the rescaled training embeddings and the OLS test projections, which is what makes supervised prediction valid. Early fusion and late fusion appear as the boundary cases of one cluster and one cluster per modality.

Load-bearing premise

The whole guarantee rests on the true dependence graph among modalities being exactly a collection of disjoint cliques: within a cluster the population affinity is at least Delta_sep, between clusters it is exactly zero, and the empirical affinity estimates converge uniformly to those population values.

Editorial extensions

If this is right

  • Users no longer need to commit a priori to early or late fusion: DAIF's boundary cases are a jointly denoising variant with one cluster and per-modality empirical-Bayes PCA with one cluster per modality, with every intermediate number of clusters selected from the data.
  • When the true dependence is a separated clique partition, the estimated clusters and priors are asymptotically harmless: the AMP iterates' state evolution matches that of an oracle that knows the true clusters and priors.
  • The recovered embeddings are Bayes-optimal in the double limit as iterations and sample size grow, so downstream prediction is built on the best possible linear-Gaussian features for this factor model.
  • The OLS test projection and the rescaled training embeddings share the same asymptotic Gaussian perturbation distribution, which justifies training a predictor on one and applying it to the other even when the test sample is small.
  • In simulations, intermediate fusion beats both full fusion and no fusion on latent reconstruction and prediction error, and the gain is largest when only a subset of modalities are dependent.

Reading between the lines

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

  • An implication the paper leaves implicit is that its clique-partition assumption is also a testable design target: for data with overlapping or partial dependence, the separation gap fails, and the consistency theorems do not cover that regime; a natural extension would replace hard clusters with soft or hierarchical affinity thresholds.
  • The per-modality signal-strength threshold suggests an immediate extension: if individual modalities are below the threshold but jointly identifiable, a spectral initialization that pools modalities before PCA could lower the effective signal threshold, a direction the paper lists as future work and one that would broaden DAIF to weak-modality settings.
  • The error-in-variables representation (3.14) is generic enough that the Cox-survival loss used on one real dataset should extend to other generalized linear or censored outcomes with the same Gaussian-perturbation correction.
  • Because the framework is modular in the dependence measure, the same pipeline should work with distance correlation or nearest-neighbor measures; the theory only needs uniform convergence of the affinity estimates, so the practical effect of measure choice is testable without changing the AMP core.
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 / 4 minor

Summary. The paper proposes DAIF, a pipeline that learns the granularity of multimodal fusion from data. It models the latent subject signatures as arising from a prior that factorizes over an unknown partition of the modalities (2.5), estimates the partition by hierarchical clustering on CKA-based dissimilarities, estimates clusterwise priors by empirical Bayes, and then applies AMP-based denoising within each estimated cluster. The main theoretical results are Theorem 6.1 (weak consistency of the estimated prior, including the estimated clusters), Theorem 6.2 (state evolution of the AMP iterates coincides with that of oracle iterates), Theorem 6.3 (Bayes-optimal recovery of the signal matrices), and Theorem 6.4 (Gaussian approximation for training and test embeddings used in downstream prediction). Simulations and two real-data applications (TEA-seq and TCGA-BRCA) are used to support the framework.

Significance. If the theoretical chain holds, the paper gives a principled interpolation between early fusion (OrchAMP) and late fusion (EB-PCA), with nonparametric clustering replacing an a priori architecture choice. The appendices contain real derivations rather than sketches: Theorem H.1 is proved by an induction on the agglomerative merges, and Theorem 6.2 transfers state evolution from Nandy and Ma (2024) once the estimated priors are consistent. However, the headline "data-adaptive granularity" claim is only established under a hard clique-independence condition, and the real-data K-selection results do not currently validate the adaptive choice. The contribution is therefore best viewed as a sound intermediate-fusion framework under well-separated cluster structure, with the practical benefit of adaptation still needing stronger evidence.

major comments (3)
  1. [Section 6.1 / Appendix H.2 (Assumption H.1)] Assumption H.1 is the load-bearing premise for Theorem 6.1, but it is not derived from the factor model (2.1)-(2.5). Under the product prior (2.5) and a characteristic kernel, between-cluster population CKA is zero by independence, so the substantive content is the within-cluster gap Delta_sep > 0. Lemma H.2 only proves convergence of empirical affinities to population affinities; it does not express Delta_sep in terms of the model parameters, nor does the paper provide a robustness statement for small positive between-cluster affinities. If the dependence structure is only approximately clique-disjoint, the population between-cluster affinity is positive, Theorem H.1's induction (which relies on an empirical affinity gap of Delta_sep/3) can fail, and then the estimator in (3.5) is maximized over the wrong coordinates. This would invalidate the oracle-equivalence basis of Theorems 6.2 and 6.3. Please provide a quantitative misspecification analysis or state the separation condition as a direct assumption with a discussion of its scope.
  2. [Sections 5.1 and 5.2 (Tables 6 and 7)] The real-data K-selection results do not support the claim that DAIF learns the optimal fusion granularity. In the TEA-seq analysis, the gap statistic selected K=1 even though K=3 had a lower test RMSE (2.6854 vs 2.6867). In the TCGA-BRCA analysis, the gap statistic selected K=2 even though K=3 had the best test C-index (0.7375 vs 0.7359). No uncertainty quantification is reported for these RMSE or C-index differences, so the observed gaps may be within noise, but as presented the data-adaptive choice is never the best-performing granularity in either real-data experiment. The paper either needs error bars or bootstrap intervals that demonstrate the selection procedure is not worse than the best fixed granularity, or the claims about data-adaptive granularity should be softened.
  3. [Section 6.2 (Theorem 6.3)] The proof of Theorem 6.3 is a single sentence: it invokes Theorem 6.2 and the techniques of Theorem 5.3 of Nandy and Ma (2024), together with fixed-point equations from (5.13) of that paper. Since Bayes-optimal recovery of the signal matrices is a headline theoretical contribution, the proof should verify that the clusterwise empirical-Bayes priors estimated from the data satisfy the required fixed-point equations, and that the limiting recursions in Theorem 6.2 converge to those fixed points. As written, this step is asserted rather than demonstrated, which makes the Bayes-optimality claim not fully supported within the manuscript.
minor comments (4)
  1. [Section 5.1] The sentence "Since the gap statistic only compares K=1,2, we separately evaluate K=3" is confusing, because the standard gap statistic can be evaluated over an arbitrary range of K; the authors should clarify whether they restricted the search and why.
  2. [Equation (3.5) and Appendix H.4] The prior class P(R^{r_k}) in (3.5) is not defined in the main text, and the likelihood L_k is only introduced in Appendix H.4. The conditions needed for MLE consistency, such as compactness or moment bounds, should be stated explicitly as part of Theorem 6.1.
  3. [Equation (3.3)] The notation \hat D_h for the estimated singular-value matrix in (3.3) is easy to confuse with the true diagonal signal matrix D_h and with the principal-component matrix D^pc_h used in the same display; a clearer notation such as \tilde D_h would improve readability.
  4. [Tables 6 and 7] The RMSE and C-index values are reported without replication or bootstrap uncertainty; given that the K-selection differences are very small, each table should indicate whether the reported differences are stable across runs or splits.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular reduction found; the data-adaptive clustering and prior consistency are proven from stated separation assumptions, and the AMP transfer theorems inherit their state-evolution statements from the authors' prior work as standalone lemmas, not as fitted inputs or renamed predictions.

full rationale

I walked the claimed derivation chain. The clustering guarantee (Theorem 6.1 together with Assumptions H.1-H.2 and Lemma H.2) is a conditional consistency statement: under a separation gap Delta_sep>0 and uniform convergence of empirical CKA affinities, the estimated partition matches the true partition and the clusterwise MLE prior converges weakly to the true product prior. This is derived from stated assumptions rather than assumed. The AMP state-evolution and Bayes-optimality results (Theorems 6.2-6.3) are imported from Nandy and Ma (2024) as lemmas: the proof of Theorem 6.2 says 'Therefore the stated limits follow from Theorem 5.1 of Nandy and Ma (2024)', and Theorem 6.3 says its proof 'follows by using Theorem 6.2 and the techniques outlined in Theorem 5.3 of Nandy and Ma (2024)'. These are self-citations with overlapping authors, but they are not circular: the cited theorems are standalone results whose assumptions (Lipschitz denoisers, smooth prior classes, BBP-type initialization conditions) do not include the present paper's clustering consistency result, and the new content of DAIF feeds into those lemmas rather than being assumed by them. No quantity fitted to the response is relabeled as a prediction: the empirical Bayes priors p-mu and p-nu are estimated from feature matrices only, the predictor is trained on the extracted training embeddings and evaluated on held-out test cells or patients. The real-data sections also honestly report that the gap-statistic choices do not always match the best test metric: on TEA-seq the gap statistic chose K=1 even though K=3 had marginally lower RMSE, and on TCGA-BRCA it chose K=2 even though K=3 had the highest test C-index. The BBP-threshold limitation stated in Section 7 is a scope condition, not a hidden input. Thus no equation reduces by construction to its own input, and no fitted parameter is passed off as a prediction. The score of 2 reflects the heavy but non-circular reliance on the authors' own prior AMP theorems.

Assumptions & free parameters 7 free parameters · 8 assumptions · 0 invented entities

The framework rests on a clustered empirical Bayes prior, which is a modeling construct rather than a new physical entity. No new particle, force, mediator, or conserved quantity is introduced. All free parameters listed are algorithmic or preprocessing choices: AMP iterations, kernel bandwidth, ranks, GMM components, number of clusters, and the CNV reduction dimension. The main entity-free risk is that the cluster structure assumption (2.5) is idealizing real dependence into cliques.

free parameters (7)
  • AMP iterations T = 10
    Fixed hyperparameter in all experiments; theory requires t going to infinity for Bayes optimality but finite T is used in practice.
  • CKA Gaussian kernel bandwidth = 1
    Default bandwidth adopted across all simulations and data examples; directly controls the affinity matrix used for clustering.
  • PCA ranks in TEA-seq = r_RNA=20, r_ATAC=15
    Chosen by scree plots rather than a formal criterion; mis-specified ranks would bias empirical Bayes prior estimation and AMP.
  • PCA ranks in TCGA-BRCA = r_RNA=10, r_Meth=8
    Chosen by scree plots for the survival analysis; same rank-sensitivity concern as TEA-seq.
  • GMM component count range = 1 to floor(3 sqrt(n))
    Prior class size selected by BIC; it is a model selection choice and affects the empirical Bayes denoisers.
  • Number of clusters K = TEA-seq K=1, TCGA K=2 (gap statistic); simulations K=2
    Fusion granularity is the central model choice; the gap statistic selects it in real data, and it is oracle-given in simulations.
  • CNV low-dimensional dimension = 5
    The five highest-variance CNV genes are selected by hand before the train/test split to fit the low-dimensional modality model.
assumptions (8)
  • standard math Standard random matrix theory for spiked models (Benaych-Georges and Nadakuditi 2012; Baik et al. 2005; Paul 2007).
    Used in Proposition H.1 to characterize empirical singular vectors as signal plus Gaussian noise.
  • domain assumption Multimodal factor model (2.1)-(2.2) with i.i.d. standard Gaussian noise and Op(1) ranks under proportional asymptotics.
    All theoretical results and the test-time OLS projection are built on this exact generative model.
  • ad hoc to paper The prior factorizes into a product over unknown clusters (2.5), i.e., dependence graph is a disjoint union of cliques.
    This is the core structural assumption that makes clusterwise empirical Bayes and AMP analysis tractable.
  • domain assumption Assumption 6.1: orthonormal latents and loadings, signal strengths above BBP threshold, and sign alignment of empirical singular vectors.
    Required for PCA initialization consistency; failure would invalidate the Gaussian perturbation representation of the principal components.
  • domain assumption Assumption 6.2: posterior-mean denoisers are uniformly Lipschitz over weak neighborhoods; verified for compact, Gaussian, log-concave, and finite Gaussian mixture priors in Proposition G.1.
    Needed to transfer consistency of estimated priors to consistency of AMP denoisers and Onsager corrections.
  • ad hoc to paper Assumptions H.1 and H.2: population CKA affinities separate clusters with gap Delta_sep, and empirical affinities converge uniformly.
    Guarantees hierarchical clustering recovers the true partition; without this, Theorem 6.1 has no basis.
  • domain assumption The response model (2.3) assumes the distribution family g and link f are known up to finite-dimensional parameters; survival analysis uses a linear Cox surrogate loss (5.1).
    DAIF's theory does not explicitly prove consistency of the Cox surrogate estimator, so the survival result relies on an additional unverified modeling step.
  • domain assumption Test observations share the same loading matrices as training and have latent factors drawn from the same prior independently of training data.
    Required for the OLS test projection and for the distributional alignment in Theorem 6.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DAIF: A Data-Driven Intermediate Fusion Framework for Multimodal Supervised Learning via Approximate Message Passing." pith.science (2026). https://pith.science/paper/77MSUUYU

@misc{pith2026260802769,
  author       = {Pith},
  title        = {Pith review of: DAIF: A Data-Driven Intermediate Fusion Framework for Multimodal Supervised Learning via Approximate Message Passing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/77MSUUYU}},
  note         = {Machine review of arXiv:2608.02769}
}
read the original abstract

Multimodal supervised learning seeks to leverage multiple heterogeneous data sources to improve predictive performance. A central challenge is determining the fusion granularity across modalities: over-integration may amplify noise while under-integration fails to exploit cross-modal dependence. Existing approaches rely on pre-specified fusion architectures, from early to late fusion, that may not adapt to the underlying dependence structure among modalities. We propose DAIF, a data adaptive intermediate fusion framework that combines random matrix theory and non-parametric dependence measures to learn fusion structure directly from data. We operate under a Bayesian multimodal factor model where the prior on the latent factors determines the cross-modal dependence. Our method clusters modalities based on estimated intermodal dependence, then performs clusterwise empirical Bayes estimation of the priors. These estimated priors are used to construct denoisers within an approximate message passing (AMP) framework, yielding denoised low-dimensional features that borrow strength across related modalities while preserving modality-specific signal. The resulting embeddings are used for downstream supervised prediction. We evaluate the framework through simulations under varying dependence structures and signal regimes, comparing against several benchmark methods, and demonstrate its practical utility on two multimodal datasets, namely a trimodal TEA-seq dataset (Swanson et al., 2021) and TCGA-BRCA dataset (Goldman et al., 2020). In the first example, we predict the expression level of a T-cell differentiation marker protein and in the second case we analyze patient survival prediction based on multimodal information. Our method competes with or outperforms the state-of-the-art techniques in both prediction problems, demonstrating its versatility across diverse supervised learning tasks.

Figures

Figures reproduced from arXiv: 2608.02769 by the authors.

Figure 1
Figure 1. Prediction of CD45RA protein expression from RNA (2,000 HVGs), normalized ATAC counts [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Thematic diagram for the DAIF pipeline finite samples. At the same time, since the clusters are inferred from the data, the pipeline provides fully data adaptive intermediate fusion. Feature extraction using Orchestrated Approximate Message Passing. Given the estimated prior, we recover the low-dimensional latent factors from tXh : h P rmsu using Orchestrated Approximate Message Passing (Nandy and Ma, 2024). The alg… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

130 extracted references · 58 canonical work pages

  1. [1]

    Optimal Spectral Algorithms for Correlated Two-view Models in High Dimensions

    Optimal Spectral Algorithms for Correlated Two-view Models in High Dimensions , author=. arXiv preprint arXiv:2605.19364 , year=

    work page Pith review arXiv

  2. [2]

    Sharp Spectral Thresholds for Multi-View Spiked Wigner Models

    Sharp Spectral Thresholds for Multi-View Spiked Wigner Models , author=. arXiv preprint arXiv:2605.19894 , year=

    work page Pith review arXiv

  3. [3]

    Nature Reviews Genetics , volume=

    Integrative omics for health and disease , author=. Nature Reviews Genetics , volume=. 2018 , publisher=

    2018

  4. [4]

    The Annals of Probability , volume =

    Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices , author =. The Annals of Probability , volume =. 2005 , publisher =

    2005

  5. [5]

    Statistica Sinica , volume =

    Paul, Debashis , title =. Statistica Sinica , volume =

  6. [6]

    2017 , journal=

    Adam: A Method for Stochastic Optimization , author=. 2017 , journal=

    2017

  7. [7]

    Comprehensive Integration of Single-Cell Data

    Stuart, Tim and Butler, Andrew and Hoffman, Paul and Hafemeister, Christoph and Papalexi, Efthymia and Mauck, 3rd, William M and Hao, Yuhan and Stoeckius, Marlon and Smibert, Peter and Satija, Rahul. Comprehensive Integration of Single-Cell Data. Cell

  8. [8]

    Mauck and Shiwei Zheng and Andrew Butler and Maddie J

    Yuhan Hao and Stephanie Hao and Erica Andersen-Nissen and William M. Mauck and Shiwei Zheng and Andrew Butler and Maddie J. Lee and Aaron J. Wilk and Charlotte Darby and Michael Zager and Paul Hoffman and Marlon Stoeckius and Efthymia Papalexi and Eleni P. Mimitou and Jaison Jain and Avi Srivastava and Tim Stuart and Lamar M. Fleming and Bertrand Yeung an...

Show all 130 references
  1. [9]

    IEEE Transactions on Information Theory , volume =

    The Dynamics of Message Passing on Dense Graphs, with Applications to Compressed Sensing , author =. IEEE Transactions on Information Theory , volume =. 2011 , month =

    2011

  2. [10]

    , title =

    Zhou, Haowen and Panwar, Pratibha and Guo, Boyi and Hallinan, Caleb and Ghazanfar, Shila and Hicks, Stephanie C. , title =. 2024 , doi =. https://www.biorxiv.org/content/early/2024/10/09/2024.10.08.615307.full.pdf , journal =

    2024

  3. [11]

    Integrated analysis of multimodal single-cell data with structural similarity

    Cao, Yingxin and Fu, Laiyi and Wu, Jie and Peng, Qinke and Nie, Qing and Zhang, Jing and Xie, Xiaohui. Integrated analysis of multimodal single-cell data with structural similarity. Nucleic Acids Res

  4. [12]

    Nature biotechnology , volume=

    Batch effects in single-cell RNA-sequencing data are corrected by matching mutual nearest neighbors , author=. Nature biotechnology , volume=. 2018 , publisher=

    2018

  5. [13]

    Liu , title =

    Zhirui Hu and Zheng Tracy Ke and Jun S. Liu , title =. Statistical Science , volume =

  6. [14]

    The Annals of Statistics , volume =

    Nonparametric Regression with Errors in Variables , author =. The Annals of Statistics , volume =

  7. [15]

    Journal of the American Statistical Association , volume=

    A design-adaptive local polynomial estimator for the errors-in-variables problem , author=. Journal of the American Statistical Association , volume=. 2009 , publisher=

    2009

  8. [16]

    The Annals of Statistics , number =

    Wenhua Jiang and Cun-Hui Zhang , title =. The Annals of Statistics , number =

  9. [17]

    Benaych-Georges, Florent and Nadakuditi, Raj Rao , title =. J. Multivar. Anal. , month = oct, pages =. 2012 , issue_date =

    2012

  10. [18]

    Molecular psychiatry , volume=

    Integrative omics of schizophrenia: from genetic determinants to clinical classification and risk prediction , author=. Molecular psychiatry , volume=. 2022 , publisher=

    2022

  11. [19]

    A kernel statistical test of independence , year =

    Gretton, Arthur and Fukumizu, Kenji and Teo, Choon Hui and Song, Le and Sch\". A kernel statistical test of independence , year =. Proceedings of the 21st International Conference on Neural Information Processing Systems , pages =

  12. [20]

    Systems biology in cardiovascular disease: a multiomics approach

    Joshi, Abhishek and Rienks, Marieke and Theofilatos, Konstantinos and Mayr, Manuel. Systems biology in cardiovascular disease: a multiomics approach. Nat Rev Cardiol

  13. [21]

    Flexynesis: A deep learning toolkit for bulk multi-omics data integration for precision oncology and beyond , volume =

    Uyar, Bora and Savchyn, Taras and Naghsh Nilchi, Amirhossein and Sarigun, Ahmet and Wurmus, Ricardo and Shaik, Mohammed Maqsood and Gr. Flexynesis: A deep learning toolkit for bulk multi-omics data integration for precision oncology and beyond , volume =. Nature Communications...

  14. [22]

    and Stegle, Oliver , title =

    Argelaguet, Ricard and Arnol, Damien and Bredikhin, Danila and Deloro, Yannik and Velten, Britta and Marioni, John C. and Stegle, Oliver , title =. Genome Biology , year =

  15. [23]

    The Annals of Applied Statistics , volume=

    Bayesian learning of clinically meaningful sepsis phenotypes in northern Tanzania , author=. The Annals of Applied Statistics , volume=

  16. [24]

    Computational Statistics & Data Analysis , volume=

    Bayesian simultaneous factorization and prediction using multi-omic data , author=. Computational Statistics & Data Analysis , volume=. 2024 , publisher=

    2024

  17. [25]

    2020 , volume =

    Method of the Year 2019: Single-cell multimodal omics , journal =. 2020 , volume =

    2019

  18. [26]

    International conference on machine learning , pages=

    Similarity of neural network representations revisited , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  19. [27]

    and Zheng, Shiwei and Butler, Andrew and Lee, Maddie J

    Hao, Yuhan and Hao, Stephanie and Andersen-Nissen, Erica and Mauck, William M. and Zheng, Shiwei and Butler, Andrew and Lee, Maddie J. and Wilk, Aaron J. and Darby, Charlotte and Zager, Michael and Hoffman, Paul and Stoeckius, Marlon and Papalexi, Efthymia and Mimitou, Eleni P...

  20. [28]

    Biometrics , volume =

    Zhang, Yunfeng and Gaynanova, Irina , title =. Biometrics , volume =. 2022 , doi =

    2022

  21. [29]

    Journal of multivariate analysis , volume=

    Angle-based joint and individual variation explained , author=. Journal of multivariate analysis , volume=. 2018 , publisher=

    2018

  22. [30]

    Bayesian Joint Additive Factor Models for Multiview Learning , journal =

    Anceschi, Niccol. Bayesian Joint Additive Factor Models for Multiview Learning , journal =. 2023 , eprint =

    2023

  23. [31]

    eLife , publisher =

    Simultaneous trimodal single-cell measurement of transcripts, epitopes, and chromatin accessibility using TEA-seq , author =. eLife , publisher =

  24. [32]

    Biometrika , volume=

    Canonical analysis of several sets of variables , author=. Biometrika , volume=. 1971 , publisher=

    1971

  25. [33]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Convex Clustering with Exemplar-Based Models , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  26. [34]

    The Annals of Mathematical Statistics , volume =

    Consistency of the maximum likelihood estimator in the presence of infinitely many incidental parameters , author =. The Annals of Mathematical Statistics , volume =

  27. [35]

    Journal of Machine Learning Research , volume=

    mvlearn: Multiview Machine Learning in Python , author=. Journal of Machine Learning Research , volume=

  28. [36]

    Nature Biotechnology , volume =

    Cao, Zhi-Jie and Gao, Guoji , title =. Nature Biotechnology , volume =. 2022 , doi =

    2022

  29. [37]

    Conference on Learning Theory , pages=

    On empirical bayes variational autoencoder: An excess risk bound , author=. Conference on Learning Theory , pages=. 2021 , organization=

    2021

  30. [38]

    arXiv preprint arXiv:2007.10389 , year=

    Generalizing Variational Autoencoders with Hierarchical Empirical Bayes , author=. arXiv preprint arXiv:2007.10389 , year=

    2007 arXiv

  31. [39]

    Journal of Machine Learning Research , volume=

    Neural empirical bayes , author=. Journal of Machine Learning Research , volume=

  32. [40]

    International Conference on Learning Representations , year =

    Clustering by Denoising: Latent plug-and-play diffusionfor single-cell embeddings , author =. International Conference on Learning Representations , year =

  33. [41]

    arXiv preprint arXiv:2502.09844 , year=

    Solving empirical bayes via transformers , author=. arXiv preprint arXiv:2502.09844 , year=

    arXiv

  34. [42]

    Stein's Unbiased Risk Estimate and Hyv

    Ghosh, Sulagna and Ignatiadis, Nikolaos and Koehler, Frederic and Lee, Amber , journal =. Stein's Unbiased Risk Estimate and Hyv. 2025 , eprint =

    2025

  35. [43]

    Nature Communications , volume =

    Zhang, Zhe and Sun, Hao and Mariappan, Rajan and others , title =. Nature Communications , volume =. 2023 , doi =

    2023

  36. [44]

    Integration and Querying of Multimodal Single-Cell Data with PoE-VAE

    Litinetskaya, Anastasia and Schulman, Maiia and Curion, Fabiola and Szalata, Artur and Omidi, Alireza and Lotfollahi, Mohammad and Theis, Fabian. Integration and Querying of Multimodal Single-Cell Data with PoE-VAE. Research in Computational Molecular Biology. 2025

    2025

  37. [45]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =

    Tibshirani, Robert and Walther, Guenther and Hastie, Trevor , title =. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =. 2001 , month =

    2001

  38. [46]

    Proceedings of the 13th annual ACM international conference on Multimedia , pages=

    Early versus late fusion in semantic video analysis , author=. Proceedings of the 13th annual ACM international conference on Multimedia , pages=

  39. [47]

    and Craft, Brian and Hastie, Mim and Repe

    Goldman, Mary J. and Craft, Brian and Hastie, Mim and Repe. Visualizing and interpreting cancer genomics data via the Xena platform , journal=. 2020 , month=

    2020

  40. [48]

    Nature , number =

    Comprehensive molecular portraits of human breast tumours , volume =. Nature , number =

  41. [49]

    Simultaneous epitope and transcriptome measurement in single cells , volume =

    Stoeckius, Marlon and Hafemeister, Christoph and Stephenson, William and Houck-Loomis, Brian and Chattopadhyay, Pratip K and Swerdlow, Harold and Satija, Rahul and Smibert, Peter , id =. Simultaneous epitope and transcriptome measurement in single cells , volume =. Nature Meth...

  42. [50]

    Lakkis, Justin and Schroeder, Amelia and Su, Kenong and Lee, Michelle Y. Y. and Bashore, Alexander C. and Reilly, Muredach P. and Li, Mingyao , journal =. A multi-use deep learning method for CITE-seq and single-cell RNA-seq data integration with cell surface protein predictio...

  43. [51]

    and Lareau, Caleb A

    Mimitou, Eleni P. and Lareau, Caleb A. and Chen, Kelvin Y. and Zorzetto-Fernandes, Andre L. and Hao, Yuhan and Takeshima, Yusuke and Luo, Wendy and Huang, Tse-Shun and Yeung, Bertrand Z. and Papalexi, Efthymia and Thakore, Pratiksha I. and Kibayashi, Tatsuya and Wing, James Ba...

  44. [52]

    Advances in neural information processing systems , volume=

    Random features for large-scale kernel machines , author=. Advances in neural information processing systems , volume=

  45. [53]

    arXiv preprint arXiv:2507.20089 , year=

    Meta Fusion: A Unified Framework For Multimodality Fusion with Mutual Learning , author=. arXiv preprint arXiv:2507.20089 , year=

    arXiv

  46. [54]

    Advances in neural information processing systems , volume=

    Multimodal generative models for scalable weakly-supervised learning , author=. Advances in neural information processing systems , volume=

  47. [55]

    The Journal of Machine Learning Research , volume=

    Multimodal learning with deep boltzmann machines , author=. The Journal of Machine Learning Research , volume=

  48. [56]

    arXiv preprint arXiv:2407.16870 , year=

    CoCA: Cooperative Component Analysis , author=. arXiv preprint arXiv:2407.16870 , year=

    arXiv

  49. [57]

    The University of Kansas Science Bulletin , volume =

    A Statistical Method for Evaluating Systematic Relationships , author =. The University of Kansas Science Bulletin , volume =

  50. [58]

    Biometrika , volume =

    Sergazinov, R and Taeb, A and Gaynanova, I , title =. Biometrika , volume =. 2026 , month =

    2026

  51. [59]

    Biometrics , volume=

    Structural learning and integrative decomposition of multi-view data , author=. Biometrics , volume=. 2019 , publisher=

    2019

  52. [60]

    arXiv preprint arXiv:2505.12473 , year=

    Multi-modal contrastive learning adapts to intrinsic dimensions of shared latent variables , author=. arXiv preprint arXiv:2505.12473 , year=

    arXiv

  53. [61]

    Multi-omics single-cell data integration and regulatory inference with graph-linked embedding , url =

    Cao, Zhi-Jie and Gao, Ge , date =. Multi-omics single-cell data integration and regulatory inference with graph-linked embedding , url =. Nature Biotechnology , number =. 2022 , bdsk-url-1 =. doi:10.1038/s41587-022-01284-4 , id =

    2022 doi

  54. [62]

    and Lin, Kevin Z

    Chen, Shuxiao and Zhu, Bokai and Huang, Sijia and Hickey, John W. and Lin, Kevin Z. and Snyder, Michael and Greenleaf, William J. and Nolan, Garry P. and Zhang, Nancy R. and Ma, Zongming , date =. Integration of spatial and single-cell data across modalities with weakly linked...

    2024 doi

  55. [63]

    arXiv preprint arXiv:2406.00778 , year=

    Bayesian joint additive factor models for multiview learning , author=. arXiv preprint arXiv:2406.00778 , year=

  56. [64]

    Psychometrika , volume =

    Levin, Jerome , title =. Psychometrika , volume =. 1966 , doi =

    1966

  57. [65]

    and Hossein‐Zadeh, G

    Afshin‐Pour, B. and Hossein‐Zadeh, G. A. and Strother, S. C. and Soltanian‐Zadeh, H. , title =. NeuroImage , volume =

  58. [66]

    , title =

    Gower, John C. , title =. Psychometrika , volume =. 1975 , doi =

    1975

  59. [67]

    Food Quality and Preference , volume =

    Pagès, Jérôme , title =. Food Quality and Preference , volume =. 2005 , doi =

    2005

  60. [68]

    arXiv preprint arXiv:1807.03748 , year=

    Representation Learning with Contrastive Predictive Coding , author=. arXiv preprint arXiv:1807.03748 , year=. 1807.03748 , archivePrefix=

    arXiv

  61. [69]

    Proceedings of the 35th International Conference on Machine Learning , series =

    MINE: Mutual Information Neural Estimation , author =. Proceedings of the 35th International Conference on Machine Learning , series =. 2018 , publisher =

    2018

  62. [70]

    The Annals of Statistics , number =

    G. The Annals of Statistics , number =

  63. [71]

    and Måge, Inger and Næs, Tormod and Hankemeier, Tom and Lips, Marjan and Kiers, Henk A

    Smilde, Age K. and Måge, Inger and Næs, Tormod and Hankemeier, Tom and Lips, Marjan and Kiers, Henk A. L. and Acar, Evrim and Bro, Rasmus , title =. Journal of Chemometrics , volume =. 2017 , doi =

    2017

  64. [72]

    and Van Mechelen, Iven , title =

    Schouteden, Maarten and Van Deun, Katrijn and Wilderjans, Tom F. and Van Mechelen, Iven , title =. Behavior Research Methods , volume =. 2014 , doi =

    2014

  65. [73]

    and Kourti, Theodora and MacGregor, J

    Westerhuis, Johan A. and Kourti, Theodora and MacGregor, J. F. , title =. Journal of Chemometrics , volume =. 1998 , doi =

    1998

  66. [74]

    2004 , isbn =

    Le Roux, Bertrand and Rouanet, Hervé , title =. 2004 , isbn =

    2004

  67. [75]

    arXiv preprint arXiv:2501.09336 , year=

    Estimating shared subspace with AJIVE: the power and limitation of multiple data matrices , author=. arXiv preprint arXiv:2501.09336 , year=

    arXiv

  68. [76]

    arXiv preprint arXiv:2507.22170 , year=

    Stacked SVD or SVD stacked? A Random Matrix Theory perspective on data integration , author=. arXiv preprint arXiv:2507.22170 , year=

    arXiv

  69. [77]

    Optimal Estimation of Shared Singular Subspaces Across Multiple Noisy Matrices , year=

    Ma, Zhengchi and Ma, Rong , journal=. Optimal Estimation of Shared Singular Subspaces Across Multiple Noisy Matrices , year=

  70. [78]

    Multi-modal contrastive learning adapts to intrinsic dimensions of shared latent variables , volume =

    Gui, Yu and Ma, Cong and Ma, Zongming , booktitle =. Multi-modal contrastive learning adapts to intrinsic dimensions of shared latent variables , volume =

  71. [79]

    Prothero, Jack and Jiang, Meilei and Hannig, Jan and Tran-Dinh, Quoc and Ackerman, Andrew and Marron, J. S. , journal =. Data integration via analysis of subspaces (DIVAS) , volume =

  72. [80]

    Proceedings of the National Academy of Sciences , volume=

    Message-passing algorithms for compressed sensing , author=. Proceedings of the National Academy of Sciences , volume=. 2009 , publisher=

    2009

  73. [81]

    The Annals of Statistics , volume =

    Estimation of Low-Rank Matrices via Approximate Message Passing , author =. The Annals of Statistics , volume =

  74. [82]

    Learning in Graphical Models , editor =

  75. [83]

    arXiv preprint arXiv:1702.00473 , year=

    Fundamental limits of low-rank matrix estimation: the non-symmetric case , author=. arXiv preprint arXiv:1702.00473 , year=

    arXiv

  76. [84]

    , title =

    Brascamp, Herm Jan and Lieb, Elliott H. , title =. Journal of Functional Analysis , volume =

  77. [85]

    Journal of the American Statistical Association , volume =

    Koenker, Roger and Mizera, Ivan , title =. Journal of the American Statistical Association , volume =

  78. [86]

    1996 , isbn =

    Graphical Models , author =. 1996 , isbn =

    1996

  79. [87]

    Information and Inference: A Journal of the IMA , volume =

    Graph-based approximate message passing iterations , author =. Information and Inference: A Journal of the IMA , volume =

  80. [88]

    arXiv preprint arXiv:2306.15580 , year=

    Approximate message passing for the matrix tensor product model , author=. arXiv preprint arXiv:2306.15580 , year=

    arXiv

  81. [89]

    The Annals of Statistics , volume =

    Fundamental limits of community detection from multi-view data: Multi-layer, dynamic and partially labeled block models , author =. The Annals of Statistics , volume =

  82. [90]

    2006 , isbn =

    Greenacre, Michael and Blasius, Jörg , title =. 2006 , isbn =

    2006

  83. [91]

    and Hoadley, Katherine A

    Lock, Eric F. and Hoadley, Katherine A. and Marron, J. Scott and Nobel, Andrew B. , title =. The Annals of Applied Statistics , volume =. 2013 , doi =

    2013

  84. [92]

    arXiv preprint arXiv:2010.01768 , year =

    Measuring Association on Topological Spaces Using Kernels and Geometric Graphs , author =. arXiv preprint arXiv:2010.01768 , year =

    2010 arXiv

  85. [93]

    Statistical Science , volume=

    Comment: Variational Autoencoders as Empirical Bayes , author=. Statistical Science , volume=. 2019 , publisher=

    2019

  86. [94]

    Briefings in Bioinformatics , volume =

    Sören Richard Stahlschmidt and Benjamin Ulfenborg and Jane Synnergren , title =. Briefings in Bioinformatics , volume =. 2022 , publisher =

    2022

  87. [95]

    Proceedings of the 2022 International Conference on Management of Data (SIGMOD '22) , year =

    Dmitrii Babaev and Nikita Ovsov and Ivan Kireev and Maria Ivanova and Gleb Gusev and Ivan Nazarov and Alexander Tuzhilin , title =. Proceedings of the 2022 International Conference on Management of Data (SIGMOD '22) , year =

    2022

  88. [96]

    arXiv preprint arXiv:2407.19030 , year=

    Multimodal data integration and cross-modal querying via orchestrated approximate message passing , author=. arXiv preprint arXiv:2407.19030 , year=

  89. [97]

    Information-theoretic limits of a multiview low-rank symmetric spiked matrix model , year=

    Barbier, Jean and Reeves, Galen , booktitle=. Information-theoretic limits of a multiview low-rank symmetric spiked matrix model , year=

  90. [98]

    Additivity of information in multilayer networks via additive Gaussian noise transforms , year=

    Reeves, Galen , booktitle=. Additivity of information in multilayer networks via additive Gaussian noise transforms , year=

  91. [99]

    Mutual Information in Community Detection with Covariate Information and Correlated Networks , year=

    Mayya, Vaishakhi and Reeves, Galen , booktitle=. Mutual Information in Community Detection with Covariate Information and Correlated Networks , year=

  92. [100]

    Fundamental Limits for High-Dimensional Factor Regression Models , year=

    Rossetti, Riccardo and Reeves, Galen , booktitle=. Fundamental Limits for High-Dimensional Factor Regression Models , year=

  93. [101]

    Community Detection With Contextual Multilayer Networks , year=

    Ma, Zongming and Nandy, Sagnik , journal=. Community Detection With Contextual Multilayer Networks , year=

  94. [102]

    The singular values and vectors of low rank perturbations of large rectangular random matrices , volume =

    Florent Benaych-Georges and Raj Rao Nadakuditi , journal =. The singular values and vectors of low rank perturbations of large rectangular random matrices , volume =

  95. [103]

    Current Bioinformatics , volume=

    A review of ensemble methods in bioinformatics , author=. Current Bioinformatics , volume=

  96. [104]

    Scientific Reports , volume=

    Learning from longitudinal data in electronic health record and genetic data to improve cardiovascular event prediction , author=. Scientific Reports , volume=. 2019 , publisher=

    2019

  97. [105]

    IEEE Transactions on Medical Imaging , year=

    Pathomic fusion: an integrated framework for fusing histopathology and genomic features for cancer diagnosis and prognosis , author=. IEEE Transactions on Medical Imaging , year=

  98. [106]

    Nature Biotechnology , volume=

    Assessing the clinical utility of cancer genomic and proteomic data across tumor types , author=. Nature Biotechnology , volume=. 2014 , publisher=

    2014

  99. [107]

    Journal of the National Cancer Institute , volume=

    Integrating tumor and stromal gene expression signatures with clinical indices for survival stratification of early-stage non--small cell lung cancer , author=. Journal of the National Cancer Institute , volume=. 2015 , publisher=

    2015

  100. [108]

    Clinical Cancer Research , volume=

    Deep learning--based multi-omics integration robustly predicts survival in liver cancer , author=. Clinical Cancer Research , volume=. 2018 , publisher=

    2018

  101. [109]

    Journal of Machine Learning Research , volume=

    Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion , author=. Journal of Machine Learning Research , volume=

  102. [110]

    Joint probabilistic modeling of single-cell multi-omic data with totalVI , volume =

    Gayoso, Adam and Steier, Zo. Joint probabilistic modeling of single-cell multi-omic data with totalVI , volume =. Nature methods , number =

  103. [111]

    Proceedings of the 30th International Conference on Machine Learning , year=

    Deep Canonical Correlation Analysis , author=. Proceedings of the 30th International Conference on Machine Learning , year=

  104. [112]

    2024 , journal=

    An Intermediate Fusion ViT Enables Efficient Text-Image Alignment in Diffusion Models , author=. 2024 , journal=

    2024

  105. [113]

    Proceedings of the National Academy of Sciences , volume=

    Cooperative learning for multiview analysis , author=. Proceedings of the National Academy of Sciences , volume=. 2022 , publisher=

    2022

  106. [114]

    and Li, Y

    Yang, X. and Li, Y. and Wang, J. and Jia, Y. and Yi, Zhang and Chen, Mao , title =. Precision Clinical Medicine , volume =. 2025 , month =

    2025

  107. [115]

    2025 , author =

    A systematic review of intermediate fusion in multimodal deep learning for biomedical applications , journal =. 2025 , author =

    2025

  108. [116]

    Machine Vision and Applications , year=

    Boulahia, Said Yacine and Amamra, Abdenour and Madi, Mohamed Ridha and Daikh, Said , title=. Machine Vision and Applications , year=

  109. [117]

    and Soffer, Shelly and Barash, Yiftach and Burshtein, Evyatar and Konen, Eli and Greenspan, Hayit , title=

    Cahan, Noa and Klang, Eyal and Marom, Edith M. and Soffer, Shelly and Barash, Yiftach and Burshtein, Evyatar and Konen, Eli and Greenspan, Hayit , title=. Scientific Reports , year=

  110. [118]

    Journal of the American Statistical Association , volume =

    Quefeng Li and Lexin Li , title =. Journal of the American Statistical Association , volume =. 2022 , publisher =. doi:10.1080/01621459.2021.1914635 , note =

    2022

  111. [119]

    Journal of the American Statistical Association , volume =

    Daolin Pang and Hongyu Zhao and Tao Wang , title =. Journal of the American Statistical Association , volume =. 2024 , publisher =. doi:10.1080/01621459.2023.2231577 , URL =

    2024

  112. [120]

    Taylor and Qi Long , title =

    Changgee Chang and Jeong Hoon Jang and Amita Manatunga and Andrew T. Taylor and Qi Long , title =. Journal of the American Statistical Association , volume =. 2020 , publisher =. doi:10.1080/01621459.2019.1689983 , URL =

    2020

  113. [121]

    Bernoulli , number =

    Xin Bing and Florentina Bunea and Marten Wegkamp , title =. Bernoulli , number =. 2022 , doi =

    2022

  114. [122]

    Journal of Machine Learning Research , year =

    Xin Bing and Florentina Bunea and Seth Strimas-Mackey and Marten Wegkamp , title =. Journal of Machine Learning Research , year =

  115. [123]

    Biometrika , volume=

    Relations Between Two Sets of Variates , author=. Biometrika , volume=

  116. [124]

    Journal of econometrics , volume=

    Factor-adjusted regularized model selection , author=. Journal of econometrics , volume=. 2020 , publisher=

    2020

  117. [125]

    Journal of the American Statistical Association , volume=

    Integrative factor regression and its inference for multimodal data analysis , author=. Journal of the American Statistical Association , volume=. 2022 , publisher=

    2022

  118. [126]

    and Tzankov, Alexandar and Lee, Ivan T

    Zhu, Bokai and Chen, Shuxiao and Bai, Yunhao and Chen, Han and Liao, Guanrui and Mukherjee, Nilanjan and Vazquez, Gustavo and McIlwain, David R. and Tzankov, Alexandar and Lee, Ivan T. and Matter, Matthias S. and Goltsev, Yury and Ma, Zongming and Nolan, Garry P. and Jiang, Si...

    2023 doi

  119. [127]

    Proceedings of the National Academy of Sciences , volume=

    Integration and transfer learning of single-cell transcriptomes via cFIT , author=. Proceedings of the National Academy of Sciences , volume=. 2021 , publisher=

    2021

  120. [128]

    Journal of the Royal Statistical Society Series B , volume =

    Empirical Bayes PCA in high dimensions , author =. Journal of the Royal Statistical Society Series B , volume =

  121. [129]

    Journal of Statistical Mechanics: Theory and Experiment , year =

    Optimal thresholds and algorithms for a model of multi-modal learning in high dimensions , author =. Journal of Statistical Mechanics: Theory and Experiment , year =

  122. [130]

    arXiv preprint arXiv:2506.02664 , year=

    Computational Thresholds in Multi-Modal Learning via the Spiked Matrix-Tensor Model , author=. arXiv preprint arXiv:2506.02664 , year=

    arXiv

Pith tools

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