Past, Present and Future of Software for Bayesian Inference

Statistical Science, Journal Year: 2024, Volume and Issue: 39(1)

Published: Feb. 1, 2024

Software tools for Bayesian inference have undergone rapid evolution in the past three decades, following popularisation of first generation MCMC-sampler implementations. More recently, exponential growth number users has been stimulated both by active development new packages machine learning community and popularity specialist software particular applications. This review aims to summarize most popular provide a useful map reader navigate world computation. We anticipate vigorous continued algorithms corresponding multiple research fields, such as probabilistic programming, likelihood-free neural networks, which will further broaden possibilities employing paradigm exciting

Language: Английский

---

Automatic cross-validation in structured models: Is it time to leave out leave-one-out?

Spatial Statistics, Journal Year: 2024, Volume and Issue: 62, P. 100843 - 100843

Published: June 12, 2024

Language: Английский

---

Big problems in spatio-temporal disease mapping: Methods and software

Computer Methods and Programs in Biomedicine, Journal Year: 2023, Volume and Issue: 231, P. 107403 - 107403

Published: Feb. 3, 2023

Fitting spatio-temporal models for areal data is crucial in many fields such as cancer epidemiology. However, when sets are very large, issues arise. The main objective of this paper to propose a general procedure analyze high-dimensional count data, with special emphasis on mortality/incidence relative risk estimation. We present pragmatic and simple idea that permits fit hierarchical the number small areas large. Model fitting carried out using integrated nested Laplace approximations over partition spatial domain. also use parallel distributed strategies speed up computations setting where Bayesian model generally prohibitively time-consuming even unfeasible. Using simulated real we show our method outperforms classical global models. implement methods algorithms develop open-source R package bigDM specific vignettes have been included facilitate methodology non-expert users. Our scalable proposal provides reliable estimates data.

Language: Английский

---

High-dimensional order-free multivariate spatial disease mapping

Statistics and Computing, Journal Year: 2023, Volume and Issue: 33(5)

Published: July 19, 2023

Despite the amount of research on disease mapping in recent years, use multivariate models for areal spatial data remains limited due to difficulties implementation and computational burden. These problems are exacerbated when number small areas is very large. In this paper, we introduce an order-free scalable Bayesian modelling approach smooth mortality (or incidence) risks several diseases simultaneously. The proposal partitions domain into smaller subregions, fits each subdivision obtains posterior distribution relative across entire domain. also provides correlations among patterns partition that combined through a consensus Monte Carlo algorithm obtain whole study region. We implement using integrated nested Laplace approximations (INLA) R package bigDM it jointly analyse colorectal, lung, stomach cancer Spanish municipalities. new permits analysis big sets better results than fitting single model.

Language: Английский

---

Fast and flexible inference for joint models of multivariate longitudinal and survival data using integrated nested Laplace approximations

Biostatistics, Journal Year: 2023, Volume and Issue: 25(2), P. 429 - 448

Published: Aug. 2, 2023

Abstract Modeling longitudinal and survival data jointly offers many advantages such as addressing measurement error missing in the processes, understanding quantifying association between markers events, predicting risk of events based on markers. A joint model involves multiple submodels (one for each longitudinal/survival outcome) usually linked together through correlated or shared random effects. Their estimation is computationally expensive (particularly due to a multidimensional integration likelihood over effects distribution) so that inference methods become rapidly intractable, restricts applications models small number and/or We introduce Bayesian approximation integrated nested Laplace algorithm implemented R package R-INLA alleviate computational burden allow multivariate with fewer restrictions. Our simulation studies show substantially reduces computation time variability parameter estimates compared alternative strategies. further apply methodology analyze five (3 continuous, 1 count, binary, 16 effects) competing risks death transplantation clinical trial primary biliary cholangitis. provides fast reliable technique applying complex encountered health research.

Language: Английский

---

A flexible Bayesian tool for CoDa mixed models: logistic-normal distribution with Dirichlet covariance

Statistics and Computing, Journal Year: 2024, Volume and Issue: 34(3)

Published: April 16, 2024

Abstract Compositional Data Analysis (CoDa) has gained popularity in recent years. This type of data consists values from disjoint categories that sum up to a constant. Both Dirichlet regression and logistic-normal have become popular as CoDa analysis methods. However, fitting this kind multivariate models presents challenges, especially when structured random effects are included the model, such temporal or spatial effects. To overcome these we propose Model (LNDM). We seamlessly incorporate approach into R-INLA package, facilitating model prediction within framework Latent Gaussian Models. Moreover, explore metrics like Deviance Information Criteria, Watanabe Akaike information criterion, cross-validation measure conditional predictive ordinate for selection CoDa. Illustrating LNDM through two simulated examples with an ecological case study on Arabidopsis thaliana Iberian Peninsula, underscore its potential effective tool managing large databases.

Language: Английский

---

A Bayesian Joint Model of Multiple Nonlinear Longitudinal and Competing Risks Outcomes for Dynamic Prediction in Multiple Myeloma: Joint Estimation and Corrected Two‐Stage Approaches

Statistics in Medicine, Journal Year: 2025, Volume and Issue: 44(3-4)

Published: Jan. 26, 2025

ABSTRACT Predicting cancer‐associated clinical events is challenging in oncology. In Multiple Myeloma (MM), a cancer of plasma cells, disease progression determined by changes biomarkers, such as serum concentration the paraprotein secreted cells (M‐protein). Therefore, time‐dependent behavior M‐protein and transition across lines therapy (LoT), which may be consequence progression, should accounted for statistical models to predict relevant outcomes. Furthermore, it important understand contribution patterns longitudinal upon each LoT initiation, time‐to‐death or time‐to‐next‐LoT. Motivated these challenges, we propose Bayesian joint model trajectories multiple M‐protein, competing risks death next LoT. Additionally, explore two estimation approaches our model: simultaneous all parameters (joint estimation) sequential using corrected two‐stage strategy aiming reduce computational time. Our proposed methods are applied retrospective cohort study from real‐world database patients diagnosed with MM US January 2015 February 2022. We split data into training test sets order validate both make dynamic predictions times until interest, informed longitudinally measured biomarkers baseline variables available up time prediction.

Language: Английский

---

A computationally efficient procedure for combining ecological datasets by means of sequential consensus inference

Environmental and Ecological Statistics, Journal Year: 2025, Volume and Issue: unknown

Published: March 11, 2025

Language: Английский

---

Parallelized integrated nested Laplace approximations for fast Bayesian inference

Statistics and Computing, Journal Year: 2022, Volume and Issue: 33(1)

Published: Dec. 24, 2022

Language: Английский

---

Mapping the abundance of endemic mosquito-borne diseases vectors in southern Quebec

BMC Public Health, Journal Year: 2023, Volume and Issue: 23(1)

Published: May 22, 2023

Abstract Background Climate change is increasing the dispersion of mosquitoes and spread viruses which some are main vectors. In Quebec, surveillance management endemic mosquito-borne diseases, such as West Nile virus or Eastern equine encephalitis, could be improved by mapping areas risk supporting vector populations. However, there currently no active tool tailored to Quebec that can predict mosquito population abundances, we propose, with this work, help fill gap. Methods Four species mosquitos were studied in project for period from 2003 2016 southern part province Quebec: Aedes vexans (VEX), Coquillettidia perturbans (CQP), Culex pipiens-restuans group (CPR) Ochlerotatus stimulans (SMG) species. We used a negative binomial regression approach, including spatial component, model abundances each function meteorological land-cover variables. tested several sets variables combination, regional local scale landcover different lag day capture weather variables, finally select one best Results Models selected showed importance independently environmental at larger scale. these models, most important predictors favored CQP VEX ‘forest’, ‘agriculture’ (for only). Land-cover ‘urban’ had impact on SMG CQP. The conditions trapping previous summarized over 30 90 days preferred shorter seven days, suggesting current long-term effects abundance. Conclusions strength component highlights difficulties modelling abundance selection shows selecting right predictors, especially when choosing temporal landscape group, it possible consider their use predicting variationsin potentially harmful public health Quebec.

Language: Английский

---