Pith. sign in

REVIEW 3 major objections 7 minor 44 references

Bayesian Probit Multi-Study Non-negative Matrix Factorization for Mutational Signatures

T0 review · 3 major / 7 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper introduces a Bayesian multi-study non-negative matrix factorization that assigns each tumor a sparse, covariate-driven set of active mutational signatures, enabling prevalence estimation and patient subtype discovery.

desk verdict A genuine methodological advance for subject-level mutational signature analysis, held back by an unvalidated variational approximation that a careful revision can fix. read the letter →

arxiv 2502.01468 v1 pith:GHZ3FFSM submitted 2025-02-03 stat.AP stat.ME

Blake Hansen , Isabella N. Grabski , Giovanni Parmigiani , Roberta De Vito This is my paper · ORCID
classification stat.APstat.ME MSC 62P1062F1562H25
keywords mutationalsignaturesBayesiannon-negativematrixfactorizationmulti-studyanalysisprobitmixturepriorsubject-levelsignatureassignmentprevalencevariationalinferencecancergenomics
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

Mutational signatures are the distinct patterns of somatic mutations left in tumor genomes by different mutagenic processes, and estimating which signatures are active in each tumor matters for understanding cancer origins and for treatment. This paper proposes a Bayesian multi-study non-negative matrix factorization that analyzes mutation counts from several cancer types at once instead of one at a time. Its core move is a per-subject binary switch for each signature, governed by a probit model on patient covariates, so a tumor's profile is explained by a sparse subset of signatures rather than by all signatures. The paper argues that this shift from study-level to subject-level signature assignment improves detection of rare signatures in simulations, yields prevalence estimates per cancer type, and reveals interpretable tumor subtypes in seven cancer types. A variational algorithm and a recovery-discovery prior that incorporates previously catalogued signatures make the approach computationally practical.

What carries the argument

Key machinery: the latent inclusion switch. For each subject $j$, signature $r$, and study $s$, the model introduces a binary variable $a_{sjr} \in \{0,1\}$ and a continuous latent score $a^*_{sjr} \sim \mathrm{N}(\beta_{sr}^\top x_{sj}, 1)$; the exposure column $\mathbf{E}_{sj}$ is given a Dirichlet prior whose concentration parameter for signature $r$ is $\alpha_{sjr} = a_{sjr}\alpha^1_s + (1-a_{sjr})\alpha^0_s$. Thus signature presence is a covariate-driven subject-level decision rather than a study-level one. A coordinate-ascent variational inference (CAVI) algorithm estimates the model, with latent Poisson allocations of each mutation to a signature, and an optional recovery-discovery prior places a Dirichlet concentrated on previously catalogued signature profiles for the known columns. The switch carries the argument: it is why prevalence can be computed as the average inclusion probability and why clusters of tumors sharing inclusion patterns can be read directly off the latent scores.

What would settle it

Run an exact Markov-chain Monte Carlo fit of the same model on a moderate simulated dataset with known per-subject signature inclusion, and compare the variational inclusion probabilities; if the 95% credible intervals exclude the true statuses noticeably more than 5% of the time, the subject-level claims fail.

Watch

Extended reading notes

Core claim

The central claim is that replacing study-level binary signature indicators with individual-level binary inclusion indicators tied to covariates through a probit link lets one multi-study NMF model simultaneously recover shared mutational signatures, estimate the prevalence of each signature within each cancer type, and identify de novo patient subtypes from the pattern of active signatures. In simulations where some signatures are active in only a small fraction of samples, the model detects those low-prevalence signatures in 94% of replicates, compared with 60-62% for the study-level Bayesian multi-study NMF, because it does not have to commit an entire study to a single presence or absence status. In real-data-inspired simulations, the recovery-discovery version attains an average cosine similarity of at least 0.99, and in the seven-cancer-type whole-genome case study it recovers known smoking- and age-associated signatures, assigns them at different prevalence levels across cancer types, and flags two candidate signatures not in the reference catalog.

Load-bearing premise

The claim stands only if the automated approximation of the statistical model produces posterior numbers that are accurate enough to trust at the level of a single patient, and the paper does not check that approximation against an exact or gold-standard calculation.

Editorial extensions

If this is right

  • Prevalence of a signature within a cancer type becomes a number that can be reported as the proportion of tumors where the signature is active, rather than a binary study-level flag.
  • Covariates such as age and sex enter the inclusion model directly, so the method can quantify different smoking-related signature prevalence in lung adenocarcinoma, lung squamous cell carcinoma, and head and neck cancer.
  • Known signatures can be fixed through a recovery-discovery prior while new signatures are still estimated, and the simulations show that this restores high accuracy in small datasets where discovery-only estimation degrades.
  • The variational computation reduces runtime from hours to minutes in the reported scenarios, making multi-study signature analysis feasible on large whole-genome datasets.

Reading between the lines

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

  • A direct extension of the paper's logic is to feed the per-tumor inclusion probabilities into downstream survival or treatment-response models, since the paper itself only clusters and estimates prevalence.
  • The same probit-switch construction should carry over to other mutation-count alphabets, such as indel classes or doublet-base profiles, where rare signatures create the same detection issue.
  • Because the case study is small, the two newly discovered signature candidates would need validation on independent cohorts before being confirmed as mutational processes; the paper itself flags this need.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper proposes BaP Multi-NMF, a Bayesian multi-study non-negative matrix factorization model for mutational signatures that moves signature presence/absence modeling from the study level to the individual tumor level. The generative model assumes Ms ~ Poisson(P Es Ws), with each exposure column Esj following a Dirichlet distribution whose concentration is driven by a binary indicator asjr, which in turn is governed by a probit regression on subject covariates, a*_sjr ~ N(beta_sr^T x_sj, 1). A recovery-discovery prior is provided that concentrates the signature prior on COSMIC catalog entries while still permitting de novo discovery. Inference is performed by coordinate-ascent variational inference, with mean-field factorization, Taylor-series expansions, and a logistic approximation to the probit link detailed in the supplement. The method is compared with Ba Multi-NMF in three simulation scenarios, with the headline result that low-prevalence signatures are detected at over 94% versus 60-62% in Scenario 2. The paper then applies the model to PCAWG data from seven cancer types and reports study-level signature prevalence, covariate associations (e.g., age with SBS1; sex with SBS4 and SBS92), subject-level clustering into interpretable groups, and two discovered signatures D1 and D2. An R package implementing the method is provided.

Significance. If validated, the contribution is meaningful: the model is a principled extension of multi-study NMF to subject-level sparse exposures with covariates, filling a real gap, since existing multi-study tools operate at the study level and rely on slower MCMC. The paper ships a fully specified generative model, self-contained CAVI derivations in the supplement, a recovery-discovery prior that plugs the COSMIC catalog in directly, a systematic comparison with the main competitor, and public R code. The principal weakness is that all subject-level claims--the abstract's promised more precise individual-level identification, prevalence estimation, and covariate-informed discovery--are transmitted through a variational posterior whose calibration is never checked; mean-field VB is known to understate posterior variance, and the multiple approximations in the algorithm make this a genuine correctness risk. The real-data discoveries (D1 and D2) are appropriately hedged by the authors' own caveat that further validation is needed, but the prevalence estimates, covariate intervals, and clustering claims inherit the same calibration risk.

major comments (3)
  1. [§3.4 and Supplementary §§1-2] The subject-level inferences that anchor the paper's contribution--the inclusion probabilities P(asjr=1), the credible intervals for beta_sr in Figures 3 and 7, and the heatmap of a*_sjr in Figure 5--all come from a variational posterior built on three stacked approximations: the mean-field factorization, the second-order Taylor expansions of expectations in the update for theta_a_srj (Supplementary Section 1.1.3), and the Bowling et al. logistic approximation to the probit link (Algorithm S1, step 7). No calibration evidence is provided: there is no MCMC benchmark (the Discussion notes that no MCMC implementation exists), no coverage assessment of the nominal 95% intervals, and no per-subject classification or calibration metric against the known asjr in the simulations; the comparisons in Figure 2 are study-level averages. Because mean-field VB generally understates posterior variance, the intervals in Figures 3 and 7 may be overconfident, and the thresholding behavior described in Section 4.3 (a_hat_sjr set to 0 for small esjr) further complicates the interpretation of theta_a_srj. I request a small-scale MCMC or importance-sampling comparison on a subset of Scenario 2 replicates, a coverage check of the beta_sr intervals, and per-subject metrics (e.g., AUROC or a calibration curve for the estimated inclusion probabilities) against the known asjr.
  2. [§4.1 and §4.3] Scenario 3, which is meant to be the most realistic assessment, cannot validate the subject-level component of the model. The ground-truth signature matrix is itself estimated: single-study frequentist NMF fits to PCAWG are matched to COSMIC at a cosine similarity threshold of 0.7 and the matched set is used as the ground truth (Section 4.1); Section 4.3 then states that no ground-truth inclusion probabilities are available for this scenario, so only average log unnormalized exposures are compared. Consequently, the most realistic setting provides no check on P(asjr=1) or on the beta_sr regression, which are exactly the quantities the method is designed to deliver. In addition, the near-equal performance of the recovery-discovery versions (avg cosine similarity >=0.99 in both methods) largely reflects the strength of an informative prior centered on the same COSMIC catalog from which the ground truth was matched. I ask the authors to add a simulation with known per-subject asjr under realistic K=96 and moderate study sizes, or at minimum to report per-subject metrics in Scenario 2, and to soften the claim in Section 4.1 that Scenario 3 served as a robust assessment of signature detection.
  3. [§3.3, §4.1, and §5] The prior hyperparameters that control sparsity and the behavior of the method in practice are not reported: the values of alpha_e0s and alpha_e1s are defined in Section 3.3 but never specified for the simulations or for the PCAWG analysis, and the recovery prior concentration c_r in equation (13) is likewise not given. The application's conclusions--the 43 retained signatures, the discovered signatures D1 and D2, and the covariate associations in Figure 7--also depend on the ad-hoc filter that keeps signatures whose 95th percentile of a* is non-negative, on the choice R_discov=5, and on the initialization, which Section 6 acknowledges as a source of sensitivity. Without a sensitivity analysis over these settings (at least the concentration parameters, the filter cutoff, and several initializations), it is difficult to know which of the reported prevalence patterns and covariate effects are robust.
minor comments (7)
  1. [Table 1] The proportion-of-male entry for Breast Adenocarcinoma reads '0.05%', which is presumably 5% (0.05); please correct the notation or the value.
  2. [§4.2] The text reports BaP Multi-NMF's Scenario 3 discovery-only average cosine similarity only as '>=0.8'; please report the actual mean and dispersion, since a value near the 0.8 capture threshold means something quite different from the 0.99 values in the other arms.
  3. [§4.3 and Figure 3] The 95% credible intervals are computed from 'the average mean and covariance of beta_sr across all 50 simulations'; this averages posterior summaries over replicates and should be explicitly described as a summary across simulations, not as the posterior interval of a single analysis.
  4. [§4.4] The runtime and memory comparison is useful, but please state the convergence criteria and the computing hardware used for both methods so that the comparison is reproducible.
  5. [§5] The filtering rule for the 91 fitted signatures (retaining signatures whose 95th percentile of a* across all subjects and cancer groups is 0 or greater) needs a one-sentence justification, since a different percentile could materially change the 43-signature list and the downstream conclusions.
  6. [References] Several entries are cited as '2023+' and '2024+' (Grabski, Trippa and Parmigiani; Zito and Miller); please update them to the published or most recent versions so that the baseline and comparison methods can be located.
  7. [Funding] The funding statement contains a typo: 'P20GM109035and' should read 'P20GM109035 and'.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity in the recovery-discovery simulation: the COSMIC prior is also the Scenario 3 ground truth, so the reported ≥0.99 cosine similarity is by construction; the core subject-level claims are independently validated.

  1. self definitional [Section 3.5, Eq. (13); Section 4.1 (Scenario 3 ground truth); Section 4.2 (recovery-discovery results)]
    "PRecov_r ∼ Dirichlet(c_r γRecov_r), where c_r ≫ 0, r = 1,...,R recov, γRecov_r is signature r of ΓRecov. ... The matched COSMIC signatures (PCOSMIC) were used as the ground truth, and unnormalized exposures (Es) were estimated by minimizing the Frobenius norm ||Ms − PCOSMIC Es||F. ... the recovery-discovery versions of both models performed equally, achieving a similarly high average cosine similarity, i.e. ≥ 0.99."

    In Scenario 3 the ground truth is the matched COSMIC signature set PCOSMIC, and Eq. (13) places the recovery prior on the same COSMIC signatures with a huge concentration c_r >> 0, so the posterior mean of each recovered signature is essentially its COSMIC prior input. The reported cosine similarity ≥0.99 is therefore a measure of how closely the posterior tracks the prior that was set to the target; it is forced by construction rather than an independent confirmation of signature estimation. The discovery-only Scenario 3 result (average cosine similarity ≥0.8) and the Scenarios 1-2 subject-level evaluations provide independent evidence, so this is a partial by-construction evaluation in a supporting simulation, not a circular central derivation.

full rationale

The core of the paper is a generative Bayesian model (Poisson NMF with probit mixture indicators) estimated by CAVI, and the central subject-level claims are tested in Scenarios 1 and 2 against ground truth simulated independently from the same model class, so those predictions are not equivalent to the model inputs. The recovery-discovery prior is an explicit informative prior on COSMIC signatures, and the only circular element is the Scenario 3 evaluation, where the same COSMIC signatures serve as both the prior input and the ground truth; with c_r >> 0 the reported cosine-similarity result is partly by construction. This does not undermine the main subject-level and covariate-effect findings, which are validated without such overlap. The self-citations to Grabski et al. are used as a baseline and for comparison, not as an unverified premise. The variational approximation and the logistic approximation to the probit link are correctness and calibration risks rather than circularity. Overall, the central derivation is self-contained, with one partial by-construction evaluation in the recovery-discovery simulation.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The model introduces several hyperparameters that are not specified in the paper, and the discovered signatures D1 and D2 are new entities without independent evidence.

free parameters (5)
  • Exposure mixture concentration parameters alpha_e0s and alpha_e1s = not specified in main text
    Chosen hyperparameters controlling the sparsity prior on exposures; values are not reported in the main text and affect how aggressively signatures are zeroed out.
  • Number of signatures R (rank) = Rrecov=86, Rdiscov=5 in application
    User-specified model dimension; not inferred from data, and signature recovery depends on it.
  • Threshold for selecting detected signatures = 95th percentile of a* >= 0
    Post-estimation filter that reduces 91 signatures to 43 recovered and 2 discovered; choice affects which signatures are reported.
  • Hyperpriors lambda_w, a_w, b_w, gamma_1, gamma_2, beta_0 = not specified in main text
    Prior parameters for gamma and normal priors; values influence the posterior and are not detailed in the paper.
  • Recovery prior concentration c_r = large constant, exact value not given
    Concentration for recovered signatures in the recovery-discovery prior (Eq. 13); must be large to fix signatures near COSMIC profiles.
assumptions (5)
  • domain assumption Observed mutation counts follow Poisson(PEsWs) with latent allocation z multinomial
    Section 3.1 Eqs (1)-(3); standard NMF likelihood for mutational counts, but not verified for all cancer types.
  • domain assumption Columns of P and Es are Dirichlet-distributed and normalized so that P Es columns sum to 1
    Section 3.2 Eqs (4)-(5); resolves scale identifiability but imposes a specific prior structure.
  • domain assumption The probit mixture on exposure concentration parameters captures signature presence
    Section 3.3 Eqs (9)-(11); assumes subject-level presence/absence is driven by covariates through a probit model.
  • ad hoc to paper Mean-field variational family with CAVI and Taylor/logistic approximations yields a reliable posterior
    Supplementary Sections 1 and 2; non-convex ELBO and approximations are not formally justified, and Section 6 acknowledges sensitivity to initialization.
  • domain assumption Recovery-discovery prior treats COSMIC signatures as fixed and correct
    Section 3.5 Eqs (12)-(13); known signatures are assumed to be exactly the COSMIC profiles with concentration c_r.
invented entities (2)
  • Discovered signature D1
    purpose: De novo mutational signature, highly prevalent in lung squamous cell carcinoma, possibly smoking-related
    Section 5.1; cosine similarity 0.64 to SBS29, no external validation; paper notes further validation is needed.
  • Discovered signature D2
    purpose: De novo mutational signature in colorectal and stomach adenocarcinoma, correlated with SBS17a and SBS17b
    Section 5.1; cosine similarity 0.76 to SBS17a, no external validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bayesian Probit Multi-Study Non-negative Matrix Factorization for Mutational Signatures." pith.science (2026). https://pith.science/paper/GHZ3FFSM

@misc{pith2026250201468,
  author       = {Pith},
  title        = {Pith review of: Bayesian Probit Multi-Study Non-negative Matrix Factorization for Mutational Signatures},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GHZ3FFSM}},
  note         = {Machine review of arXiv:2502.01468}
}
read the original abstract

Mutational signatures are patterns of somatic mutations in tumor genomes that provide insights into underlying mutagenic processes and cancer origin. Developing reliable methods for their estimation is of growing importance in cancer biology. Somatic mutation data are often collected for different cancer types, highlighting the need for multi-study approaches that enable joint analysis in a principled and integrative manner. Despite significant advancements, statistical models tailored for analyzing the genomes of multiple cancer types remain underexplored. In this work, we introduce a Bayesian Multi-Study Non-negative Matrix Factorization (NMF) approach that uses mixture modeling to incorporate sparsity in the exposure weights of each subject to mutational signatures, allowing for individual tumor profiles to be represented by a subset rather than all signatures, and making this subset depend on covariates. This allows for a) more precise ability to identify meaningful contributions of mutational signatures at the individual level; b) estimation of the prevalence of activity of signatures within a cancer type, defined by the proportion of tumor profiles where a certain signature is present; and c) de-novo identification of interpretable patient subtypes based on the mutational signatures present within their mutational profile. We apply our approach to the mutational profiles of tumors from seven different cancer types, demonstrating its ability to accurately estimate mutational signatures while uncovering both individual and tissue-specific differences. An R package implementing our method is available at https://github.com/blhansen/BAPmultiNMF.

Figures

Figures reproduced from arXiv: 2502.01468 by the authors.

Figure 7
Figure 7. Visualization of study-specific covariate effects, reported as a 95% credible interval based on the variational approximation of BaP Multi-NMF. Due to limitations noted in Section 2, we excluded the sex covariate for the breast adenocarcinoma samples, and we excluded both covariates for esophagus adenocarcinoma. consistent with the known higher prevalence of tobacco use among men (Israel T. et al., 2014). Furthermor… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 42 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION format.url url empty "" url if FUNCTION article output.bibitem format.authors "author" output.check author format.key output output.year.check new.block format.title "title" output.check new.block crossref missing format.jour.vol output format.article.crossref output.nonnull format.pages output if ne...

  2. [2]

    barticle [author] Aaltonen , Lauri A. L. A. , Abascal , Federico F. , Abeshouse , Adam A. The ICGC/TCGA Pan-Cancer Analysis of Whole Genomes Consortium ( 2020 ). Pan-cancer analysis of whole genomes . Nature 578 82--93 . barticle

    work page 2020

  3. [3]

    barticle [author] Alexandrov , Ludmil B. L. B. , Nik-Zainal , Serena S. , Wedge , David C. D. C. , Aparicio , Samuel A. J. R. S. A. J. R. et al. ( 2013 ). Signatures of mutational processes in human cancer . Nature 500 415--421 . barticle

    work page 2013

  4. [4]

    barticle [author] Alexandrov , Ludmil B. L. B. , Jones , Philip H. P. H. , Wedge , David C. D. C. , Sale , Julian E. J. E. et al. ( 2015 ). Clock-like mutational processes in human somatic cells . Nature Genetics 47 . barticle

    work page 2015

  5. [5]

    barticle [author] Alexandrov , Ludmil B. L. B. , Kim , Jaegil J. , Haradhvala , Nicholas J. N. J. , Huang , Mi Ni M. N. et al. ( 2020 ). The repertoire of mutational signatures in human cancer . Nature 578 94--101 . barticle

    work page 2020

  6. [6]

    bbook [author] Bishop , Christopher M. C. M. ( 2006 ). Pattern Recognition and Machine Learning . Information science and statistics . Springer , New York . bbook

    work page 2006

  7. [7]

    barticle [author] Chandra , Noirrit Kiran N. K. , Dunson , David B. D. B. Xu , Jason J. ( 2024 ). Inferring Covariance Structure from Multiple Data Sources via Subspace Factor Analysis . Journal of the American Statistical Association 0 1--15 . barticle

    work page 2024

  8. [8]

    barticle [author] Dahlstrom , Kristina R. K. R. , Little , Jarrod A. J. A. , Zafereo , Mark E. M. E. , Lung , Margaret M. et al. ( 2008 ). Squamous cell carcinoma of the head and neck in never smoker–never drinkers: A descriptive epidemiologic study . Head & Neck 30 75--84 . barticle

    work page 2008

Show all 44 references
  1. [9]

    , Bellio , Ruggero R

    barticle [author] De Vito , Roberta R. , Bellio , Ruggero R. , Trippa , Lorenzo L. Parmigiani , Giovanni G. ( 2019 ). Multi‐study factor analysis . Biometrics 75 337--346 . barticle

    2019

  2. [10]

    barticle [author] Goldman , Mary J. M. J. , Craft , Brian B. , Hastie , Mim M. , Repečka , Kristupas K. et al. ( 2020 ). Visualizing and interpreting cancer genomics data via the Xena platform . Nature Biotechnology 38 675--678 . barticle

    2020

  3. [11]

    barticle [author] Grabski , Isabella N. I. N. , Trippa , Lorenzo L. Parmigiani , Giovanni G. ( 2023 +). Bayesian Multi - Study Non - Negative Matrix Factorization for Mutational Signatures . bioRxiv:2023.03.28.534619 . barticle

    2023

  4. [12]

    barticle [author] Grabski , Isabella N I. N. , De Vito , Roberta R. , Trippa , Lorenzo L. Parmigiani , Giovanni G. ( 2023 ). Bayesian Combinatorial Multi-Study Factor Analysis . The Annals of Applied Statistics 17 2212 . barticle

    2023

  5. [13]

    barticle [author] Gulhan , Doga C. D. C. , Lee , Jake June-Koo J. J.-K. , Melloni , Giorgio E. M. G. E. M. , Cort \'e s-Ciriano , Isidro I. Park , Peter J. P. J. ( 2019 ). Detecting the mutational signature of homologous recombination deficiency in clinical samples . Nature Ge...

    2019

  6. [14]

    , Avalos - Pacheco , Alejandra A

    barticle [author] Hansen , Blake B. , Avalos - Pacheco , Alejandra A. , Russo , Massimiliano M. Vito , Roberta De R. D. ( 2024 ). Fast Variational Inference for Bayesian Factor Analysis in Single and Multi-Study Settings . Journal of Computational and Graphical Statistics 0 1-...

    2024

  7. [15]

    barticle [author] Hayward , Nicholas K. N. K. , Wilmott , James S. J. S. , Waddell , Nicola N. , Johansson , Peter A. P. A. et al. ( 2017 ). Whole-genome landscapes of major melanoma subtypes . Nature 545 175--180 . barticle

    2017

  8. [16]

    , Eshtad , Saeed S

    barticle [author] Helleday , Thomas T. , Eshtad , Saeed S. Nik-Zainal , Serena S. ( 2014 ). Mechanisms underlying mutational signatures in human cancers . Nature Reviews Genetics 15 585--598 . barticle

    2014

  9. [17]

    barticle [author] Islam , S. M. Ashiqul S. M. A. , Díaz-Gay , Marcos M. , Wu , Yang Y. , Barnes , Mark M. et al. ( 2022 ). Uncovering novel mutational signatures by de novo extraction with SigProfilerExtractor . Cell Genomics 2 . barticle

    2022

  10. [18]

    , Agaku A

    barticle [author] Israel T. , Agaku A. , King , Brian A. B. A. , Husten , Corinne G. C. G. , Bunnell , Rebecca R. et al. ( 2014 ). Tobacco Product Use Among Adults — United States , 2012–2013 . Morbidity and Mortality Weekly Report 63 542--547 . barticle

    2014

  11. [19]

    , Gulhan , Doga C

    barticle [author] Jin , Hu H. , Gulhan , Doga C. D. C. , Geiger , Benedikt B. , Ben-Isvy , Daniel D. et al. ( 2024 ). Accurate and sensitive mutational signature analysis with MuSiCal . Nature Genetics 56 541--552 . barticle

    2024

  12. [20]

    , Degasperi , Andrea A

    barticle [author] Koh , Gene G. , Degasperi , Andrea A. , Zou , Xueqing X. , Momen , Sophie S. et al. ( 2021 ). Mutational signatures: emerging concepts, caveats and clinical applications . Nature Reviews Cancer 21 619--637 . barticle

    2021

  13. [21]

    barticle [author] Lawson , Andrew R. J. A. R. J. , Abascal , Federico F. , Coorens , Tim H. H. T. H. H. , Hooks , Yvette Y. et al. ( 2020 ). Extensive heterogeneity in somatic mutation and selection in the human bladder . Science 370 75--82 . barticle

    2020

  14. [22]

    barticle [author] Loeb , Keith R. K. R. Loeb , Lawrence A. L. A. ( 2000 ). Significance of multiple mutations in cancer . Carcinogenesis 21 379--385 . barticle

    2000

  15. [23]

    , Alexandrov , Ludmil B

    barticle [author] Nik-Zainal , Serena S. , Alexandrov , Ludmil B. L. B. , Wedge , David C. D. C. , Van Loo , Peter P. et al. ( 2012 ). Mutational Processes Molding the Genomes of 21 Breast Cancers . Cell 149 979--993 . barticle

    2012

  16. [24]

    , Rao , Vinayak V

    binproceedings [author] Petralia , Francesca F. , Rao , Vinayak V. Dunson , David D. ( 2012 ). Repulsive Mixtures . In Advances in Neural Information Processing Systems ( F. F. Pereira , C. J. C. J. Burges , L. L. Bottou K. Q. K. Q. Weinberger , eds.) 25 . Curran Associates, I...

    2012

  17. [25]

    ( 2017 )

    barticle [author] Quinn , Thomas T. ( 2017 ). peak RAM : Monitor the T otal and P eak RAM U sed by an E xpression or F unction . R package version 1.0.2 . barticle

    2017

  18. [26]

    bphdthesis [author] Roberts , Nicola Diane N. D. ( 2018 ). Patterns of somatic genome rearrangement in human cancer , Thesis , University of Cambridge . bphdthesis

    2018

  19. [27]

    barticle [author] Rosales , Rafael A R. A. , Drummond , Rodrigo D R. D. , Valieris , Renan R. , Dias-Neto , Emmanuel E. et al. ( 2017 ). signeR : an empirical Bayesian approach to mutational signature discovery . Bioinformatics 33 8--16 . barticle

    2017

  20. [28]

    , Lavine , Isaac I

    barticle [author] Roy , Arkaprava A. , Lavine , Isaac I. , Herring , Amy H. A. H. Dunson , David B. D. B. ( 2021 ). Perturbed factor analysis: Accounting for group differences in exposure profiles . The Annals of Applied Statistics 15 1386--1404 . barticle

    2021

  21. [29]

    barticle [author] Samur , Mehmet Kemal M. K. , Aktas Samur , Anil A. , Fulciniti , Mariateresa M. , Szalat , Raphael R. et al. ( 2020 ). Genome- Wide Somatic Alterations in Multiple Myeloma Reveal a Superior Outcome Group . Journal of Clinical Oncology 38 3107--3118 . barticle

    2020

  22. [30]

    , Li , Xiaodun X

    barticle [author] Secrier , Maria M. , Li , Xiaodun X. , de Silva , Nadeera N. , Eldridge , Matthew D. M. D. et al. ( 2016 ). Mutational signatures in esophageal adenocarcinoma define etiologically distinct subgroups with therapeutic relevance . Nature Genetics 48 1131--1141 ....

    2016

  23. [31]

    , Moody , Sarah S

    barticle [author] Senkin , Sergey S. , Moody , Sarah S. , Díaz-Gay , Marcos M. , Abedi-Ardekani , Behnoush B. et al. ( 2024 ). Geographic variation of mutagenic exposures in kidney cancer genomes . Nature 629 910--918 . barticle

    2024

  24. [32]

    barticle [author] Stratton , Michael R. M. R. ( 2011 ). Exploring the Genomes of Cancer Cells : Progress and Promise . Science 331 1553--1558 . barticle

    2011

  25. [33]

    barticle [author] Stratton , Michael R. M. R. , Campbell , Peter J. P. J. Futreal , P. Andrew P. A. ( 2009 ). The cancer genome . Nature 458 719--724 . barticle

    2009

  26. [34]

    Govindan , Ramaswamy R

    barticle [author] Subramanian , Janakiraman J. Govindan , Ramaswamy R. ( 2007 ). Lung Cancer in Never Smokers : A Review . Journal of Clinical Oncology 25 561--570 . barticle

    2007

  27. [35]

    barticle [author] Tate , John G J. G. , Bamford , Sally S. , Jubb , Harry C H. C. , Sondka , Zbyslaw Z. et al. ( 2019 ). COSMIC : the Catalogue Of Somatic Mutations In Cancer . Nucleic Acids Research 47 D941--D947 . barticle

    2019

  28. [36]

    , Vogelstein , Bert B

    barticle [author] Tomasetti , Cristian C. , Vogelstein , Bert B. Parmigiani , Giovanni G. ( 2013 ). Half or more of the somatic mutations in cancers of self-renewing tissues originate prior to tumor initiation. Proc Natl Acad Sci U S A 110(6) 1999-2004 . barticle

    2013

  29. [37]

    Kinzler , Kenneth W K

    barticle [author] Vogelstein , Bert B. Kinzler , Kenneth W K. W. ( 2004 ). Cancer genes and the pathways they control. Nature Medicine 10 789--799 . barticle

    2004

  30. [38]

    , Kurutmaz , M

    barticle [author] Yıldırım , Sinan S. , Kurutmaz , M. Burak M. B. , Barsbey , Melih M. , Şimşekli , Umut U. Cemgil , A. Taylan A. T. ( 2021 ). Bayesian Allocation Model : Marginal Likelihood - Based Model Selection for Count Tensors . IEEE Journal of Selected Topics in Signal ...

    2021

  31. [39]

    Miller , Jeffrey W

    bmisc [author] Zito , Alessandro A. Miller , Jeffrey W. J. W. ( 2024 +). Compressive Bayesian non-negative matrix factorization for mutational signatures analysis . arXiv:2404.10974 [stat] . bmisc

    2024 arXiv

  32. [40]

    Bishop, C. M. (2006). Pattern Recognition and Machine Learning . Information science and statistics. Springer, New York

    2006

  33. [41]

    R., Khasawneh, M

    Bowling, S. R., Khasawneh, M. T., Kaewkuekool, S., and Cho, B. R. (2009). A logistic approximation to the cumulative normal distribution. Journal of Industrial Engineering and Management , 2(1):114--127

    2009

  34. [42]

    Gaujoux, R., Seoighe, C., and Sauwen, N. (2024). The package NMF: manual pages . R package version 0.28

    2024

  35. [43]

    Ishiguro, K., Sato, I., and Ueda, N. (2017). Averaged Collapsed Variational Bayes Inference . Journal of Machine Learning Research , 18(1):1--29

    2017

  36. [44]

    G., Hahn, T., Bouvier, A., and Kiêu, K

    Narasimhan, B., Johnson, S. G., Hahn, T., Bouvier, A., and Kiêu, K. (2024). cubature: Adaptive Multivariate Integration over Hypercubes . R package version 2.1.1

    2024

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.