pith. sign in

arxiv: 2604.14364 · v1 · submitted 2026-04-15 · 📊 stat.AP

Joint Bayesian Inference of Genetic Effect Sizes and PK Parameters in Nonlinear Mixed-Effects Models

Julien Martinelli , Ibtissem Rebai , David W. Haas , Julie Bertrand This is my paper

Pith reviewed 2026-05-10 11:39 UTC · model grok-4.3

classification 📊 stat.AP
keywords Bayesian variable selectionnonlinear mixed effectspharmacokineticsSNP associationsparsity priorsjoint inferencepharmacogeneticsuncertainty quantification
0
0 comments X

The pith

A joint Bayesian model simultaneously estimates nonlinear mixed-effects PK parameters and selects relevant genetic variants with coherent posterior uncertainties.

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

The paper develops a single-stage Bayesian method that estimates both population pharmacokinetic parameters and the effects of many genetic markers together in one model. This avoids the usual two-step process where genetic selection happens separately from parameter fitting, which can ignore important uncertainties. By testing five different sparsity priors on simulated data and a real study of HIV patients on nevirapine, the approach recovers true genetic signals with low false positives and gives consistent pharmacokinetic estimates across priors. A sympathetic reader would care because pharmacogenetic analyses often suffer from small sample sizes and correlated markers, making reliable identification of relevant genes difficult without proper uncertainty handling.

Core claim

The authors show that embedding sparsity-inducing priors directly into nonlinear mixed-effects models allows joint posterior inference over pharmacokinetic parameters and SNP inclusion, yielding low false-discovery rates in simulations and stable inference on real genetic-PK data from 129 Cambodian patients.

What carries the argument

The joint posterior over NLMEM parameters and SNP effect sizes under sparsity priors such as spike-and-slab or regularized horseshoe, which performs simultaneous regularization and uncertainty quantification.

If this is right

  • Posterior inclusion probabilities for SNPs become directly available from the model fit.
  • Pharmacokinetic parameter estimates remain stable regardless of which sparsity prior is chosen.
  • The method recovers causal genetic signals with F1 scores around 0.8 in simulations while keeping false discoveries near zero.
  • Real-data posterior predictive checks confirm adequate model calibration for nevirapine PK in HIV patients.

Where Pith is reading between the lines

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

  • This framework could be extended to other nonlinear models in pharmacodynamics or disease progression.
  • Larger multi-center studies might reveal whether prior-sensitive SNPs indicate true weak effects or noise.
  • Integration with clinical outcome models could link identified variants directly to dosing recommendations.

Load-bearing premise

That the sparsity priors can be calibrated via effect-size and sparsity targets to achieve reliable recovery and low false-discovery rates in high-dimensional SNP data without the joint model becoming too sensitive or computationally unstable.

What would settle it

Finding that posterior inclusion probabilities for the same SNPs vary dramatically across the five priors on the real ANRS 12154 dataset, or that simulated data with known causal SNPs yields high false positives under all priors.

Figures

Figures reproduced from arXiv: 2604.14364 by David W. Haas, Ibtissem Rebai, Julie Bertrand, Julien Martinelli.

Figure 1
Figure 1. Figure 1: Observed temporal concentration profiles by genotype under [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Marginal prior distributions for a given SNP effect size [PITH_FULL_IMAGE:figures/full_fig_p016_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: F1-score as a function of the posterior inclusion probability (PIP) threshold. Top panel: comparison of all five sparsity-inducing priors at fixed ε = 0.01. Bottom row: one panel per prior, showing F1 curves across multiple values of ε, except for the spike-and-slab, where analytical PIPs were used. Shaded regions denote standard errors computed over 100 datasets. 100 200 300 0.000 0.005 0.010 V1 1 2 0 1 k… view at source ↗
Figure 4
Figure 4. Figure 4: Posterior distributions of pharmacokinetic parameters across the five sparsity-inducing priors. Shown are the [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Normalized prediction distribution (NPD) visual predictive checks by sampling scheme. The sparsity [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Posterior inclusion probabilities across sparsity-inducing priors for the real data case study. [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Posterior distributions of the top five SNP effects per sparsity-inducing prior in the real data case study. For [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
read the original abstract

High-dimensional genetic covariate selection in population pharmacokinetic (PK) models is challenging due to the cohort's restricted size and high correlation among single-nucleotide polymorphisms (SNPs). We propose a fully Bayesian, single-stage framework that jointly infers nonlinear mixed effect model (NLMEM) parameters and SNP effect sizes, providing coherent posterior uncertainty and inclusion summaries within a single model fit. We compare five sparsity-inducing priors -- Spike-and-Slab, Hierarchical Lasso, Regularized Horseshoe, R2--D2, and the $\ell_1$-ball -- calibrated through effect-size and sparsity targets. In simulations, all priors showed low false-discovery rates around $0$--$0.08$ under the null, and recovered the causal signal under the alternative, with peak $F_1$ scores around $0.8$--$0.85$ under reasonable inclusion cutoffs. Spike-and-Slab was especially attractive because it provides analytical posterior inclusion probabilities directly, while among priors requiring tolerance-based proxy inclusion summaries, the $\ell_1$-ball combined similarly strong recovery with the most stable behavior across tolerance values. On genetic and PK data from the ANRS 12154 study in 129 Cambodians living with HIV and receiving nevirapine, posterior predictive checks indicated adequate calibration and PK parameter inference remained stable across priors. While the dominant signal was robust across priors, additional candidate SNPs showed only partial agreement in ranking and more prior-sensitive effect-size estimates. These results support Bayesian variable selection within joint NLMEM as a principled approach for pharmacogenetic analyses when uncertainty quantification and regularization are central.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The manuscript proposes a fully Bayesian single-stage framework for jointly inferring nonlinear mixed-effects model (NLMEM) parameters and high-dimensional SNP effect sizes in pharmacogenetics. It compares five sparsity-inducing priors (Spike-and-Slab, Hierarchical Lasso, Regularized Horseshoe, R2-D2, ℓ1-ball) calibrated to effect-size and sparsity targets. Simulations report low FDR (0–0.08) and peak F1 scores (0.8–0.85), while the ANRS 12154 application (n=129) shows stable PK inference and posterior predictive checks but only partial agreement on additional SNPs across priors.

Significance. If the joint model delivers reliable variable selection with coherent uncertainty, this would advance pharmacogenetic PK analyses by avoiding two-stage biases and providing principled regularization in small-cohort, high-dimensional settings. Strengths include the explicit comparison of multiple calibrated priors, use of posterior inclusion summaries, and real-data stability for dominant signals. The approach is particularly relevant when uncertainty quantification is central, though real-data sensitivity tempers the immediate impact.

major comments (2)
  1. [Simulation studies] Simulation studies: The reported FDR of 0–0.08 and F1 scores of ~0.8–0.85 do not appear to incorporate realistic SNP linkage disequilibrium structures or the full nonlinear mixed-effects likelihood, which are load-bearing for assessing whether prior calibration to effect-size/sparsity targets ensures stable recovery and low false-discovery rates on actual pharmacogenetic data.
  2. [ANRS 12154 analysis] ANRS 12154 analysis: The dominant signal is robust, but partial agreement in SNP ranking and prior-sensitive effect-size estimates for additional candidates (despite calibration) indicate that the five sparsity priors remain sensitive in correlated high-dimensional settings; this directly challenges the central claim of reliable joint inference and coherent posterior inclusion summaries.
minor comments (3)
  1. [Methods] Clarify the exact implementation of tolerance-based proxy inclusion summaries for the four priors other than Spike-and-Slab, including how tolerances were chosen and their impact on reported F1 scores.
  2. The abstract and results would benefit from explicitly stating the number of SNPs tested and the precise form of the NLMEM used in the real-data example to improve reproducibility.
  3. [Results] Posterior predictive check details (e.g., specific statistics and how they were computed under the joint model) should be expanded to allow readers to assess calibration adequacy.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We address each major comment point by point below, clarifying our simulation design and the interpretation of the real-data results while acknowledging limitations where appropriate.

read point-by-point responses

  1. Referee: Simulation studies: The reported FDR of 0–0.08 and F1 scores of ~0.8–0.85 do not appear to incorporate realistic SNP linkage disequilibrium structures or the full nonlinear mixed-effects likelihood, which are load-bearing for assessing whether prior calibration to effect-size/sparsity targets ensures stable recovery and low false-discovery rates on actual pharmacogenetic data.

    Authors: We appreciate the referee's emphasis on simulation realism. Our simulations were intentionally designed as a controlled evaluation to isolate the impact of the five sparsity-inducing priors under the joint Bayesian NLMEM framework, generating data from the full nonlinear mixed-effects likelihood with a known causal SNP subset (under both null and alternative scenarios). This allowed direct assessment of prior calibration to effect-size and sparsity targets without confounding by external genetic correlation structures. We acknowledge that realistic LD patterns among SNPs would constitute a more stringent test for FDR control in correlated high-dimensional settings. In the revised manuscript we will add an explicit limitations paragraph noting this design choice and recommending that future extensions incorporate LD blocks derived from real pharmacogenetic panels (e.g., via simulation tools such as HAPGEN2). The current results nevertheless establish a baseline performance for the joint model and priors; the ANRS 12154 analysis, which necessarily includes empirical LD, provides complementary evidence of practical behavior. revision: partial

  2. Referee: ANRS 12154 analysis: The dominant signal is robust, but partial agreement in SNP ranking and prior-sensitive effect-size estimates for additional candidates (despite calibration) indicate that the five sparsity priors remain sensitive in correlated high-dimensional settings; this directly challenges the central claim of reliable joint inference and coherent posterior inclusion summaries.

    Authors: We agree that the observed partial agreement on secondary SNPs illustrates the inherent difficulty of high-dimensional selection in small cohorts with correlated covariates, and our manuscript already states this explicitly. However, this does not challenge the central claim. The joint Bayesian framework's primary contribution is the delivery of coherent posterior uncertainty and inclusion summaries from a single model fit, avoiding two-stage biases. The dominant signal (consistent with prior pharmacogenetic knowledge for nevirapine) remains stable across all five priors, while variability among weaker candidates is expected and is itself quantified by the posterior inclusion probabilities (directly available for the spike-and-slab prior). We interpret the multi-prior comparison as a strength that reveals sensitivity rather than a weakness; in practice, investigators can use the range of posterior summaries to guide cautious interpretation or prioritize signals for replication. We will expand the discussion to clarify this nuance and to suggest ensemble or replication strategies for additional candidates, but no substantive change to the reported results or claim is required. revision: no

Circularity Check

0 steps flagged

No circularity: joint posterior derived directly from likelihood and priors

full rationale

The derivation consists of specifying a hierarchical Bayesian NLMEM with SNP covariates under five sparsity priors, then sampling the joint posterior. All reported quantities (parameter estimates, inclusion probabilities, posterior predictive checks) are direct functionals of that posterior. Simulations and the ANRS 12154 analysis are external validation steps that compare recovered signals against known truth or across priors; they do not redefine or tautologically reproduce the fitted quantities. No equation reduces by construction to a pre-fitted input, and no load-bearing uniqueness claim rests on self-citation. The framework is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The framework rests on standard NLMEM assumptions plus sparsity priors whose hyperparameters are calibrated to external targets; no new entities are postulated.

free parameters (1)
  • effect-size and sparsity targets
    Priors are calibrated through effect-size and sparsity targets chosen to control FDR and recovery.
axioms (2)
  • domain assumption Nonlinear mixed-effects model assumptions hold for the pharmacokinetic data.
    The paper applies NLMEM without additional validation of its structural assumptions.
  • domain assumption SNP effects are sparse.
    Sparsity-inducing priors are used, assuming only a small number of SNPs have non-zero effects.

pith-pipeline@v0.9.0 · 5600 in / 1357 out tokens · 49133 ms · 2026-05-10T11:39:47.090356+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages

  1. [1]

    Pharmacogenomics in the clinic

    Relling MV , Evans WE. Pharmacogenomics in the clinic. Nature. 2015;526(7573):343-50

    work page 2015

  2. [2]

    Comparison of nonlinear mixed effects models and noncompartmental approaches in detecting pharmacogenetic covariates

    Tessier A, Bertrand J, Chenel M, Comets E. Comparison of nonlinear mixed effects models and noncompartmental approaches in detecting pharmacogenetic covariates. The AAPS Journal. 2015;17(3):597-608

    work page 2015

  3. [3]

    Covariate selection in pharmacometric analyses: a review of methods

    Hutmacher MM, Kowalski KG. Covariate selection in pharmacometric analyses: a review of methods. British Journal of Clinical Pharmacology. 2015;79(1):132-47

    work page 2015

  4. [4]

    Bayesian high-dimensional covariate selection in non-linear mixed-effects models using the SAEM algorithm

    Naveau M, Kon Kam King G, Rincent R, Sansonnet L, Delattre M. Bayesian high-dimensional covariate selection in non-linear mixed-effects models using the SAEM algorithm. Statistics and Computing. 2024;34(1):53

    work page 2024

  5. [5]

    A Stability-Enhanced Lasso Approach for Covariate Selection in Non-Linear Mixed Effects Model

    Gabaut A, Thiébaut R, Proust-Lima C, Prague M. A Stability-Enhanced Lasso Approach for Covariate Selection in Non-Linear Mixed Effects Model. Statistics in Medicine. 2026;45(1-2):e70407

    work page 2026

  6. [6]

    Estimation and variable selection in high dimension in nonlinear mixed-effects models

    Caillebotte A, Kuhn E, Lemler S. Estimation and variable selection in high dimension in nonlinear mixed-effects models. arXiv preprint arXiv:250320401. 2025

    work page 2025

  7. [7]

    An embedded genus-one helicoid.Pro- ceedings of the National Academy of Sciences, 102(46):16566–16568, 2005.doi:10.1073/pnas

    Taylor J, Tibshirani RJ. Statistical learning and selective inference. Proceedings of the National Academy of Sciences. 2015;112(25):7629-34. Available from: https://www.pnas.org/doi/abs/10.1073/pnas. 1507583112

    work page doi:10.1073/pnas 2015

  8. [8]

    Assessing parameter uncertainty in small-n pharmacometric analyses: value of the log-likelihood profiling-based sampling importance resampling (LLP-SIR) technique

    Broeker A, Wicha SG. Assessing parameter uncertainty in small-n pharmacometric analyses: value of the log-likelihood profiling-based sampling importance resampling (LLP-SIR) technique. Journal of pharmacokinetics and pharmacodynamics. 2020;47(3):219-28

    work page 2020

  9. [9]

    Some alternatives to asymptotic tests for the analysis of pharmacoge- netic data using nonlinear mixed effects models

    Bertrand J, Comets E, Chenel M, Mentré F. Some alternatives to asymptotic tests for the analysis of pharmacoge- netic data using nonlinear mixed effects models. Biometrics. 2012;68(1):146-55

    work page 2012

  10. [10]

    PKBugs: a program for Bayesian analysis of population pharmacokinetic and pharmacodynamic models

    Lunn DJ, Best N, Thomas A, Wakefield J, Spiegelhalter D. PKBugs: a program for Bayesian analysis of population pharmacokinetic and pharmacodynamic models. Journal of Pharmacokinetics and Pharmacodynamics. 2002;29(3):239-68

    work page 2002

  11. [11]

    MCMC using Hamiltonian dynamics

    Neal RM, et al. MCMC using Hamiltonian dynamics. Handbook of markov chain monte carlo. 2011;2(11):2

    work page 2011

  12. [12]

    The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo

    Hoffman MD, Gelman A. The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research. 2014;15(47):1593-623. Available from: http://jmlr.org/papers/ v15/hoffman14a.html

    work page 2014

  13. [13]

    Pyro: Deep Universal Probabilistic Programming

    Bingham E, Chen JP, Jankowiak M, Obermeyer F, Pradhan N, Karaletsos T, et al. Pyro: Deep Universal Probabilistic Programming. J Mach Learn Res. 2019;20:28:1-28:6. Available from: http://jmlr.org/papers/ v20/18-403.html

    work page 2019

  14. [14]

    Composable Effects for Flexible and Accelerated Probabilistic Programming in NumPyro

    Phan D, Pradhan N, Jankowiak M. Composable Effects for Flexible and Accelerated Probabilistic Programming in NumPyro. arXiv preprint arXiv:191211554. 2019

    work page 2019

  15. [15]

    JAX: composable transformations of Python+NumPy programs; 2018

    Bradbury J, Frostig R, Hawkins P, Johnson MJ, Leary C, Maclaurin D, et al.. JAX: composable transformations of Python+NumPy programs; 2018. Available from:http://github.com/jax-ml/jax

    work page 2018

  16. [16]

    Dose-finding methods for Phase I clinical trials using pharmacokinetics in small populations

    Ursino M, Zohar S, Lentz F, Alberti C, Friede T, Stallard N, et al. Dose-finding methods for Phase I clinical trials using pharmacokinetics in small populations. Biometrical Journal. 2017;59(4):804-25

    work page 2017

  17. [17]

    Bayesian inference using Hamiltonian Monte-Carlo algorithm for nonlinear joint modeling in the context of cancer immunotherapy

    Kerioui M, Mercier F, Bertrand J, Tardivon C, Bruno R, Guedj J, et al. Bayesian inference using Hamiltonian Monte-Carlo algorithm for nonlinear joint modeling in the context of cancer immunotherapy. Statistics in Medicine. 2020;39(30):4853-68. 13

    work page 2020

  18. [18]

    Uncertainty Computation at Finite Distance in Non- linear Mixed Effects Models—a New Method Based on Metropolis-Hastings Algorithm

    Guhl M, Bertrand J, Fayette L, Mercier F, Comets E. Uncertainty Computation at Finite Distance in Non- linear Mixed Effects Models—a New Method Based on Metropolis-Hastings Algorithm. The AAPS Journal. 2024;26(3):53

    work page 2024

  19. [19]

    Variable Selection via Gibbs Sampling

    George EI, McCulloch RE. Variable Selection via Gibbs Sampling. Journal of the American Statistical Association. 1993;88(423):881-9

    work page 1993

  20. [20]

    Bayesian inference with the l1-ball prior: solving combinatorial problems with exact zeros

    Xu M, Duan LL. Bayesian inference with the l1-ball prior: solving combinatorial problems with exact zeros. Journal of the Royal Statistical Society Series B: Statistical Methodology. 2023 07;85(5):1538-60. Available from: https://doi.org/10.1093/jrsssb/qkad076

    work page doi:10.1093/jrsssb/qkad076 2023

  21. [21]

    Hierarchical Bayesian survival analysis and projective covariate selection in cardiovascular event risk prediction

    Peltola T, Havulinna A, Salomaa V , Vehtari A. Hierarchical Bayesian survival analysis and projective covariate selection in cardiovascular event risk prediction. CEUR Workshop Proceedings. 2014 01;1218:79-88

    work page 2014

  22. [22]

    Sparsity information and regularization in the horseshoe and other shrinkage priors

    Piironen J, Vehtari A. Sparsity information and regularization in the horseshoe and other shrinkage priors. Elec- tronic Journal of Statistics. 2017 jan;11(2). Available from:https://doi.org/10.1214%2F17-ejs1337si

    work page 2017

  23. [23]

    Bayesian Regression Using a Prior on the Model Fit: The R2-D2 Shrinkage Prior

    Zhang YD, Naughton BP, Bondell HD, Reich BJ. Bayesian Regression Using a Prior on the Model Fit: The R2-D2 Shrinkage Prior. Journal of the American Statistical Association. 2022;117(538):862-74

    work page 2022

  24. [24]

    The Regularized Horseshoe for Covariate Selection Improves Convenience and Predictive Performance in Population PK/PD Models

    Pourzanjani A, Davis C. The Regularized Horseshoe for Covariate Selection Improves Convenience and Predictive Performance in Population PK/PD Models. CPT: Pharmacometrics & Systems Pharmacology. 2026;15:e70198

    work page 2026

  25. [25]

    Is the One-Compartment Model with First Order Absorption a Useful Approximation? Pharmaceutical Research

    Weiss M. Is the One-Compartment Model with First Order Absorption a Useful Approximation? Pharmaceutical Research. 2023;40(9):2147-53

    work page 2023

  26. [26]

    Multiple genetic variants predict steady-state nevirapine clearance in HIV-infected Cambodians

    Bertrand J, Chou M, Richardson DM, Verstuyft C, Léger PD, Mentré F, et al. Multiple genetic variants predict steady-state nevirapine clearance in HIV-infected Cambodians. Pharmacogenetics and Genomics. 2012;22(12):868- 76

    work page 2012

  27. [27]

    Optimal predictive model selection

    Barbieri MM, Berger JO. Optimal predictive model selection. The Annals of Statistics. 2004;32(3):870 897. Available from:https://doi.org/10.1214/009053604000000238

    work page doi:10.1214/009053604000000238 2004

  28. [28]

    Bayesian variable selection regression for genome-wide association studies and other large-scale problems

    Guan Y , Stephens M. Bayesian variable selection regression for genome-wide association studies and other large-scale problems. The Annals of Applied Statistics. 2011;5(3):1780 1815. Available from: https://doi. org/10.1214/11-AOAS455

    work page doi:10.1214/11-aoas455 2011

  29. [29]

    A Simple New Approach to Variable Selection in Regression, with Application to Genetic Fine Mapping

    Wang G, Sarkar A, Carbonetto P, Stephens M. A Simple New Approach to Variable Selection in Regression, with Application to Genetic Fine Mapping. Journal of the Royal Statistical Society Series B: Statistical Methodology. 2020 07;82(5):1273-300. Available from:https://doi.org/10.1111/rssb.12388

    work page doi:10.1111/rssb.12388 2020

  30. [30]

    Population pharmacokinetic- pharmacogenetic study of nevirapine in HIV-infected Cambodian patients

    Chou M, Bertrand J, Segeral O, Verstuyft C, Borand L, Comets E, et al. Population pharmacokinetic- pharmacogenetic study of nevirapine in HIV-infected Cambodian patients. Antimicrobial Agents and Chemother- apy. 2010 October;54(10):4432-9

    work page 2010

  31. [31]

    Model Evaluation of Con- tinuous Data Pharmacometric Models: Metrics and Graphics

    Nguyen THT, Mouksassi MS, Holford N, Al-Huniti N, Freedman I, Hooker AC, et al. Model Evaluation of Con- tinuous Data Pharmacometric Models: Metrics and Graphics. CPT: Pharmacometrics & Systems Pharmacology. 2017;6(2):87-109. Available from: https://ascpt.onlinelibrary.wiley.com/doi/abs/10.1002/psp4. 12161

    work page doi:10.1002/psp4 2017

  32. [32]

    Variational inference: A review for statisticians

    Blei DM, Kucukelbir A, McAuliffe JD. Variational inference: A review for statisticians. Journal of the American statistical Association. 2017;112(518):859-77

    work page 2017

  33. [33]

    Group Spike-and-Slab Variational Bayes

    Komodromos M, Evangelou M, Filippi S, Ray K. Group Spike-and-Slab Variational Bayes. Bayesian Analysis. 2025;1(1):1-29

    work page 2025

  34. [34]

    Bayesian Variable Selection in a Million Dimensions

    Jankowiak M. Bayesian Variable Selection in a Million Dimensions. In: Ruiz F, Dy J, van de Meent JW, editors. Proceedings of The 26th International Conference on Artificial Intelligence and Statistics. vol. 206 of Proceedings of Machine Learning Research. PMLR; 2023. p. 253-82

    work page 2023

  35. [35]

    Increasing certainty in systems biology models using Bayesian multimodel inference

    Linden-Santangeli N, Zhang J, Kramer B, Rangamani P. Increasing certainty in systems biology models using Bayesian multimodel inference. Nature Communications. 2025;16(1):7416

    work page 2025

  36. [36]

    Variational Inference for Dirichlet Process Mixtures

    Blei DM, Jordan MI. Variational inference for Dirichlet process mixtures. Bayesian Analysis. 2006;1(1):121 143. Available from:https://doi.org/10.1214/06-BA104. 14 5 10 15H0 rs37 = 0 (N = 147) Time (hours) rs37 = 1 (N = 219) rs37 = 2 (N = 34) 1 4 12 5 10 15H1 1 4 12 Time (hours) 1 4 12 Concentration (arbitrary units) Figure 1: Observed temporal concentrat...

    work page doi:10.1214/06-ba104 2006

  37. [37]

    Orthogonality gives x⊤ j rj =x ⊤ j  y− X k̸=j xkβk   =x ⊤ j y,(S6) sincex ⊤ j xk = 0fork̸=j

    and, with standardization, ∥xj∥2 2 =N . Orthogonality gives x⊤ j rj =x ⊤ j  y− X k̸=j xkβk   =x ⊤ j y,(S6) sincex ⊤ j xk = 0fork̸=j. Plugging back, ∥y−Xβ∥ 2 2 =∥r j∥2 2 −2β jx⊤ j y+β 2 j N.(S7) Complete the square inβ j. Define the least-squares coefficient ˆβj := x⊤ j y ∥xj∥2 2 = x⊤ j y N .(S8) Then − 1 2ω2 h β2 j N−2β jx⊤ j y =− N 2ω2 h (βj − ˆβj)2 ...

    work page

  38. [38]

    theta " , dist . Beta ( gamma_a , gamma_b ) ) 10 11# SNP - wise in cl usi on i n d i c a t o r s z_j are m a r g i n a l i z e d in parallel 12with numpyro . plate (

    show that this is well approximated by an affine transform of the classical horseshoe (HS) shrinkage: ˜κj ≈(1−b)κ j +bwithb= 1 1 +N c2 .(S22) Intuitively, the slab shifts the shrinkage profile upward from(0,1) to (bj,1) , preventing shrinkage for very large signals. Then, the prior effective size rescales as ˜meff, RHS ≈(1−b)m eff, HS,(S23) 24 meaning tha...

    work page

  39. [39]

    r1 _g lo ba l

    expand ([ nSNP ]) ) 16 17# Step 3: global s hr in kag e c o m p o n e n t s 18r 1_ gl ob al = numpyro . sample ( " r1 _g lo ba l " , dist . Normal (0 , 1) ) 19r 2_ gl ob al = numpyro . sample ( " r2 _g lo ba l " , 20dist . I n v e r s e G a m m a (0.5 * nu_global , 210.5 * n u_ gl oba l ) ) 22 23# Step 4: slab c om po ne nt 24caux = numpyro . sample ( " c...

    work page 2000