The DOI matches its Crossref record; 51 of the 70 references it cites were found in a registry or an index; 13 of the 25 sources its claims cite were checked for support.
Similarity-based multimodal regression
Open accessPublished version of a preprint
Abstract
Summary: To better understand complex human phenotypes, large-scale studies have increasingly collected multiple data modalities across domains such as imaging, mobile health, and physical activity. The properties of each data type often differ substantially and require either separate analyses or extensive processing to obtain comparable features for a combined analysis. Multimodal data fusion enables certain analyses on matrix-valued and vector-valued data, but it generally cannot integrate modalities of different dimensions and data structures. For a single data modality, multivariate distance matrix regression provides a distance-based framework for regression accommodating a wide range of data types. However, no distance-based method exists to handle multiple complementary types of data. We propose a novel distance-based regression model, which we refer to as Similarity-based Multimodal Regression (SiMMR), that enables simultaneous regression of multiple modalities through their distance profiles. We demonstrate through simulation, imaging studies, and longitudinal mobile health analyses that our proposed method can detect associations between clinical variables and multimodal data of differing properties and dimensionalities, even with modest sample sizes. We perform experiments to evaluate several different test statistics and provide recommendations for applying our method across a broad range of scenarios.
Overview
CoChat Summary
TL;DR
The paper proposes similarity-based multimodal regression (SiMMR), a distance-based framework that jointly regresses multiple modalities of arbitrary data structure on covariates through their dissimilarity matrices and classical multidimensional scaling scores. Across simulations and applications to PNC neuroimaging and NIMH mobile health data, SiMMR detects associations with modest sample sizes and often outperforms unimodal or multiple-comparison distance-based alternatives.
Key findings
- SiMMR offers two test statistics: SiMMR-D, a Dempster trace equivalent to MDMR on the sum of dissimilarity matrices, and SiMMR-PC, Pillai’s trace on the top K principal components of concatenated cMDS scores; both use permutation testing with 999 permutations.
- In simulations, type I error was well-controlled for SiMMR statistics, SiMMR outperformed MC-MDMR, and SiMMR-D and KMR produced similar power; SiMMR-PC generally outperformed SiMMR-D when intermodal correlation was high and the covariate effect was sufficiently complex.
- In PNC connectivity data at sample size 40, SiMMR-D and KMR had rejection rates of 90.7% and 92.5%, compared with 83.1% for MDMR on structural connectivity and 82.9% for MC-MDMR.
- In the NIMH Family Study at sample size 50, SiMMR-PC(3) had a rejection rate of 83.2%, versus 57.1% for the highest unimodal MDMR analysis (EMA-measured irritability) and 26.2% for MC-MDMR; in PNC cortical structure at sample size 20, MDMR on cortical thickness had 83.3% rejection, while MC-MDMR, KMR, and SiMMR-D had 75.9%, 71.5%, and 72.9%, respectively, and MDMR on sulcal depth had 13.0%.
Methods
- SiMMR computes a dissimilarity matrix for each modality, doubly centers and weights each matrix by its largest eigenvalue, represents each via classical multidimensional scaling, concatenates the cMDS scores, and tests the joint association between the concatenated scores and covariates using either SiMMR-D or SiMMR-PC.
- Simulations varied sample size, number of modalities, number of features per modality, covariate effect rank, and correlation structure (exchangeable with ρ = 0.25 or 0.75, and AR(1) with τ = 0.9), generating 1,000 multimodal datasets per setting and comparing SiMMR with MDMR, MC-MDMR, KMR, and MMR where applicable.
- Applications used PNC cortical thickness and sulcal depth (n = 912), PNC structural and functional connectivity (n = 727), and NIMH Family Study actigraphy and EMA data (n = 77), with rejection rates assessed by 1,000 resamples without replacement.
Limitations
- The paper states that SiMMR’s benefit is limited when modalities show lesser correlation, that multimodal tests decreased in performance as more unassociated modalities were added, and that modalities with weak associations can reduce power even when intermodal correlation is known to be important.
- The paper notes that distance choices were not fully examined and that further investigation may suggest alternative data-driven approaches for choosing the optimal number of PCs; KMR could not be applied with categorical outcomes or non-Euclidean dissimilarity metrics.
- Evident from the text: the NIMH application analyzed only 77 subjects with at least 1 week of data from a larger subset of 384 participants with actigraphy and EMA data, and the PNC analyses used subsets (912 and 727) of the larger cohort.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
13 Valid · 5 Not valid · 11 Not checked · 86 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“We choose to compute dissimilarities between functional connectivity matrices using the log-Euclidean distance (Arsigny et al. 2006), which addresses several issues with using Frobenius distances for positive semidefinite matrices.”
What the sources say
- Partly supportedcites Arsigny et al. 2006 sourceExperimental — accuracy not yet measured
“The Defects of Euclidean Calculus” identifies problems with Frobenius-distance-based tensor calculations, and says log-Euclidean metrics preserve positive-definiteness. But the Conclusions state that the framework assumes positive-definite tensors; the claim’s broader reference to positive semidefinite matrices includes matrices with zero eigenvalues, which the paper does not cover. (LLM, conf=0.88)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In particular, SiMMR-PC resembles PC-based test statistics in the multiple phenotype setting with multiple outcome variables and a single covariate (Liu and Lin 2019); but these statistics do not apply to our settings with multiple covariates.”
What the sources say
- Partly supportedcites Liu and Lin 2019 sourceExperimental — accuracy not yet measured
The Principal Component Association Tests section describes testing one SNP at a time across multiple phenotypes, which supports the distinction from a test of multiple predictors jointly. But it also explicitly says the Z-scores account for other covariates included in the single-phenotype analyses. Thus the paper does not support the unqualified claim that its statistics cannot apply when multiple covariates are present; no page number is available in our copy. (LLM, conf=0.84)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“We also observe through our application to PNC cortical structure that modalities with weak associations can reduce power of a multimodal analysis, even when the correlation between modalities is known to be important (Vandekar et al. 2016).”
What the sources say
- Supportedcites Vandekar et al. 2016 source
The Introduction describes a strong relationship between cortical thickness and sulcal depth, and the Methods explain the value of analyzing that relationship directly. The sentence presents the reduced-power finding as the citing authors’ own observation; the citation supports the importance of the relationship between measures, not that power finding. Page numbers were not available in our copy. (LLM, conf=0.86)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We demonstrate through simulation, imaging studies, and longitudinal mobile health analyses that our proposed method can detect associations between clinical variables and multimodal data of differing properties and dimensionalities, even with modest sample sizes.”
boostedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We find that the type I error rates for SiMMR test statistics are well-controlled across simulation settings, but MC-MDMR is overly conservative at high numbers of modalities M , especially in the exchangable correlation settings.”
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
65 more claims
Create a free account to see all 115 claims and to verify the 2 references still open.
- Probably not a substantial claim. 31
- It may state two separate claims as one. 18
- The quote only partly states this claim. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.3. Selection of SiMMR-PC · p. 14
“Figure 4b shows in our PNC connectivity study that SiMMR-D and SiMMR-PC show comparable performance at varying numbers of PCs. However, Fig. 4c shows that SiMMR-PC(3) and SiMMR-PC(4) have higher power for detection of diagnosis than SiMMR-D at higher sample sizes in the NIMH Family Study application.”
§ 4.2.3. Results · p. 14
“(a) shows the rejection rate across resamples for SiMMR-PC test statistics across number of PCs compared to SiMMR-D (dashed line). (b) displays the distance correlation (DistCor) among modalities in each application using the full sample. (c) shows the percent of variation explained by PCs across the 1,000 resamplings of size 50 in each application.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.3. Selection of SiMMR-PC · p. 14
“Figure 4a shows that the distance correlation between PNC cortical thickness and sulcal depth is relatively low (0.39)”
§ 4.2.3. Results · p. 14
“(b) displays the distance correlation (DistCor) among modalities in each application using the full sample.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.1. Data generation · p. 7
“L =⌊ MQ / 4 ⌋ has c 1 = c 2 =··· = c ⌊ MQ / 4 ⌋ = 0.7, L =⌊ MQ / 4 ⌋ has c 1 = c 2 =··· = c ⌊ MQ / 2 ⌋ = 0.7”
§ 3.1. Data generation · p. 7
“We choose L as 1, ⌊ MQ / 4 ⌋ , ⌊ MQ / 2 ⌋ , and MQ to provide simulation settings with varying complexity of covariate effects.”
Citations
Cites
Advances in Neural Information Processing Systems 19
(2007) · The MIT Press eBooks · doi:10.52202/085713 · Cited by 15,539
A new method for non-parametric multivariate analysis of variance
Anderson, M. J. (2001) · Austral Ecology · doi:10.1111/j.1442-9993.2001.01070.pp.x · Cited by 13,828
Sudlow, C., Gallacher, J., Allen, N., Beral, V., Burton, P., Danesh, J., Downey, P., Elliott, P., Green, J., Landray, M., et al. (2015) · PLOS Medicine · doi:10.1371/journal.pmed.1001779 · Cited by 12,728
Cortical Surface-Based Analysis
Dale, A. M., Fischl, B., & Sereno, M. I. (1999) · NeuroImage · doi:10.1006/nimg.1998.0395 · Cited by 10,910
Cortical Surface-Based Analysis
Fischl, B., Sereno, M. I., & Dale, A. M. (1999) · NeuroImage · doi:10.1006/nimg.1998.0396 · Cited by 6,147
FITTING MULTIVARIATE MODELS TO COMMUNITY DATA: A COMMENT ON DISTANCE-BASED REDUNDANCY ANALYSIS
McArdle, B. H., & Anderson, M. J. (2001) · Ecology · doi:10.1890/0012-9658(2001)082[0290:fmmtcd]2.0.co;2 · Cited by 4,001
Least Squares Support Vector Machines
Suykens, J. A. K., Van Gestel, T., De Brabanter, J., De Moor, B., & Vandewalle, J. (2002) · WORLD SCIENTIFIC eBooks · doi:10.1142/5089 · Cited by 3,642
Measuring and testing dependence by correlation of distances
Székely, G. J., Rizzo, M. L., & Bakirov, N. K. (2007) · The Annals of Statistics · doi:10.1214/009053607000000505 · Cited by 2,684
A Kernel Method for the Two-Sample-Problem
Gretton, A., Borgwardt, K. M., Rasch, M., Schölkopf, B., & Smola, A. J. (2007) · Advances in Neural Information Processing Systems 19 · doi:10.7551/mitpress/7503.003.0069 · Cited by 2,167
Multimodal Data Fusion: An Overview of Methods, Challenges, and Prospects
Lahat, D., Adali, T., & Jutten, C. (2015) · Proceedings of the IEEE · doi:10.1109/jproc.2015.2460697 · Cited by 1,296
Cited by
Chen, A. A., Clark, K. A., Dewey, B. E., DuVal, A. L., Pellegrini, N., Nair, G., Jalkh, Y., Khalil, S., Zurawski, J., Calabresi, P. A., et al. (2024) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1012241 · Cited by 4
Self-citationMulti-modal and multi-region distance model for neuroimaging: Application to ABCD studyPreprint
Zhang, X., Vandekar, S., Chen, A. F., Kang, K., Seidlitz, J., Alexander‐Bloch, A. F., & Liu, J. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.01.21.700689 · Cited by 1
Self-citationAhmad, A., Diaz, F. J., Sendra, S., & Lloret, J. (2025) · Food Analytical Methods · doi:10.1007/s12161-025-02923-7 · Cited by 2
Husky, M. M., Hunt, L. C., Leroux, A., Dey, D., Conway, K. P., Vetter, C., Zipunnikov, V., Choudhari, R., Kass, M., Leyden, M., et al. (2025) · doi:10.2196/preprints.85667 · Cited by 1
Self-citationLiu, J., Xu, K., Ferguson, J. F., Kang, K., Wang, Y., Qiu, Y., Shao, L., Tu, S., Nguyen, T. T., Lin, T., et al. (2026) · Statistics in Medicine · doi:10.1002/sim.70654
Metadata
- Authors
- Chen, A. A., Weinstein, S. M., Adebimpe, A., Gur, R. C., Gur, R. E., Merikangas, K. R., Satterthwaite, T. D., Shinohara, R. T., & Shou, H.
- Affiliations
- Medical University of South Carolina (US)
- Temple University (US)
- University of Pennsylvania (US)
- Children's Hospital of Philadelphia (US)
- National Institute of Mental Health (US)
- Corresponding author
- Chen, Andrew A
- ORCID
- Venue
- Biostatistics
- Publisher
- Oxford University Press (OUP)
- Published
- 2023-12-06
- Volume
- 25
- Issue
- 4
- Pages
- 1122-1139
- ISSN
- 1465-4644 (ISSN-L)
- 1468-4357
- PMC
- PMC11471965
- OpenAlex
- W4389451563
- Cited by
- 6
- References
- 70
- Citations by year
- 2026: 2 · 2025: 3 · 2024: 1
- FWCI
- 0.93
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- public-domain
- Locations
- Biostatistics (publishedVersion, public-domain)
- bioRxiv (Cold Spring Harbor Laboratory) (submittedVersion, cc-by-nc-nd)
- PubMed Central (publishedVersion, public-domain)
- PubMed (publishedVersion)
- Funders
- National Institute of Neurological Disorders and Stroke (R01 NS085211, R01 NS060910)
- National Multiple Sclerosis Society (RG-1707-28586)
- National Institute of Mental Health (R01 MH123550)
- University of Pennsylvania Center for Biomedical Image Computing and Analytics
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Applications
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Statistical Methods and Models
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Inference
- Keywords
- multimodal regression, multimodal fusion, matrix-valued data, vector-valued data, mobile health, physical activity, test statistics
- MeSH terms
- Data Interpretation, Statistical, Humans, Regression Analysis, Models, Statistical
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Similarity-based multimodal regression. https://app.cochat.ai/papers/10.1093/biostatistics/kxad033
- The paper
- Chen, A. A., Weinstein, S. M., Adebimpe, A., Gur, R. C., Gur, R. E., Merikangas, K. R., Satterthwaite, T. D., Shinohara, R. T., & Shou, H. (2024). Similarity-based multimodal regression. Biostatistics. https://doi.org/10.1093/biostatistics/kxad033
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.