Bayesian Mixture Research Articles

Bayesian typhoon precipitation prediction with a mixture of ensemble forecast-based and historical event-based prediction functions

Statistical post-processing has been widely adopted in probabilistic forecasting, especially when a predictive distribution of weather variables is of interest. Among the available post-processing methods, a common tool is the ensemble Bayesian model averaging. Challenges arise when it is applied to heavy rain prediction, including the precipitation produced by tropical cyclones. The issues are related to the rare extreme event and limited available data for model training. To address this, we propose in this research a Bayesian mixture model, Bmix, based on information from forecast members and historical tropical cyclones, rather than solely on immediately observed precipitation data. This approach provides a grid-specific Bayesian predictive distribution and an easy-to-derive probabilistic forecast with posterior samples generated from the Markov chain Monte Carlo algorithm. This post-processing procedure can be executed on each grid simultaneously, reducing the computational time significantly. In addition, a categorized Bmix (C.Bmix) is implemented to accommodate different scenarios such as heavy or torrential rain in applications. The proposed approaches are demonstrated on the 24-hour accumulated precipitation forecast of a category 4-equivalent super typhoon Dujuan landing in Taiwan in 2015; while two others, Matmo and Soudelor, are adopted to provide historical information. Over the Taiwan main island, a total of 1282 grids are considered, each with a resolution of 5 km and 20 ensemble member forecasts from the Weather Research and Forecasting Ensemble Prediction System (WEPS) at a 6-hour interval during the typhoon period. The analyses show that the proposed mixture approaches present larger percentages of positive CRPSS values than traditional BMA. For instance, at the time when Dujuan made landfall, the percentages are 92.2 % for Bmix, 95.2 % for C.Bmix, and 15.4 % for BMA when the true precipitation amount exceeds 200 mm/24 h. The proposed methods outperform in the probability prediction and the overall pattern of precipitation.

Bayesian Gaussian Mixture Models for Enhanced Radar Sensor Modeling: A Data-Driven Approach towards Sensor Simulation for ADAS/AD Development.

In the realm of road safety and the evolution toward automated driving, Advanced Driver Assistance and Automated Driving (ADAS/AD) systems play a pivotal role. As the complexity of these systems grows, comprehensive testing becomes imperative, with virtual test environments becoming crucial, especially for handling diverse and challenging scenarios. Radar sensors are integral to ADAS/AD units and are known for their robust performance even in adverse conditions. However, accurately modeling the radar's perception, particularly the radar cross-section (RCS), proves challenging. This paper adopts a data-driven approach, using Gaussian mixture models (GMMs) to model the radar's perception for various vehicles and aspect angles. A Bayesian variational approach automatically infers model complexity. The model is expanded into a comprehensive radar sensor model based on object lists, incorporating occlusion effects and RCS-based detectability decisions. The model's effectiveness is demonstrated through accurate reproduction of the RCS behavior and scatter point distribution. The full capabilities of the sensor model are demonstrated in different scenarios. The flexible and modular framework has proven apt for modeling specific aspects and allows for an easy model extension. Simultaneously, alongside model extension, more extensive validation is proposed to refine accuracy and broaden the model's applicability.

Confounder-dependent Bayesian mixture model: Characterizing heterogeneity of causal effects in air pollution epidemiology.

Several epidemiological studies have provided evidence that long-term exposure to fine particulate matter (pm2.5) increases mortality rate. Furthermore, some population characteristics (e.g., age, race, and socioeconomic status) might play a crucial role in understanding vulnerability to air pollution. To inform policy, it is necessary to identify groups of the population that are more or less vulnerable to air pollution. In causal inference literature, the group average treatment effect (GATE) is a distinctive facet of the conditional average treatment effect. This widely employed metric serves to characterize the heterogeneity of a treatment effect based on some population characteristics. In this paper, we introduce a novel Confounder-Dependent Bayesian Mixture Model (CDBMM) to characterize causal effect heterogeneity. More specifically, our method leverages the flexibility of the dependent Dirichlet process to model the distribution of the potential outcomes conditionally to the covariates and the treatment levels, thus enabling us to: (i) identify heterogeneous and mutually exclusive population groups defined by similar GATEs in a data-driven way, and (ii) estimate and characterize the causal effects within each of the identified groups. Through simulations, we demonstrate the effectiveness of our method in uncovering key insights about treatment effects heterogeneity. We apply our method to claims data from Medicare enrollees in Texas. We found six mutually exclusive groups where the causal effects of pm2.5 on mortality rate are heterogeneous.

Dynamic Knowledge Injection for AIXI Agents

Prior approximations of AIXI, a Bayesian optimality notion for general reinforcement learning, can only approximate AIXI's Bayesian environment model using an a-priori defined set of models. This is a fundamental source of epistemic uncertainty for the agent in settings where the existence of systematic bias in the predefined model class cannot be resolved by simply collecting more data from the environment. We address this issue in the context of Human-AI teaming by considering a setup where additional knowledge for the agent in the form of new candidate models arrives from a human operator in an online fashion. We introduce a new agent called DynamicHedgeAIXI that maintains an exact Bayesian mixture over dynamically changing sets of models via a time-adaptive prior constructed from a variant of the Hedge algorithm. The DynamicHedgeAIXI agent is the richest direct approximation of AIXI known to date and comes with good performance guarantees. Experimental results on epidemic control on contact networks validates the agent's practical utility.

Abstract 1254: Integration of human omics analysis and new syngeneic tumor organoid models reveals that aberrant RAS/PI3K crosstalk triggers refractoriness in high-grade serous carcinoma

Abstract Refractory high-grade serous carcinoma (HGSC) is one of the most severe clinical problems in gynecology, but we have barely grasped the underlying mechanism of aggressiveness and chemo-resistance. Previous study has shown that homologous recombination genes, BRCA1/2-unmutated HGSC patients possess the potential to be in refractory condition, and their primary response to chemotherapy was heterogenous; thus, there can be subtype-specific determinant factors hatching refractoriness in BRCA1/2-unmutated HGSC patients. To explore the determinant factors, we performed gene set variation analysis and comprehensive log-rank tests on TCGA advanced serous ovarian cancer datasets. As a result, regulation of small GTPase signaling prominently correlated with poor prognosis uniquely in BRCA1/2-unmutated HGSC patients (p-value < 0.001), and eventually, pathway clustering analysis by variational Bayesian Gaussian mixture model discovered aberrant RAS/PI3K crosstalk as peculiar poor prognostic signature of BRCA1/2-unmutated HGSC patients. To elucidate the mechanism of poor progression triggered by RAS/PI3K, we utilized our syngeneic HGSC organoid models newly established from murine fallopian tube epithelium by making Myc overexpressing, and knocking out Trp53 and Rb1; as a supplement, the propriety of these genetic alterations to develop HGSC models was backed up by genetic profiles of high-grade serous ovarian cancer patients in AACR Project GENIE. After that, we knocked out Nf1 and Pten whose deletions were detected relatively with high frequency in HGSC to induce aberrant RAS/PI3K crosstalk in the primitive HGSC model. In-vivo evaluation and growth comparison indicated that aberrant RAS/PI3K crosstalk transformed Brca1/2-unmutated HGSC modeling cells into aggressive phenotypes such as poor survival (p-value < 0.01) and high growth rate (p-value < 0.001). Chemo-sensitivity assay also showed that the genetic manipulation made Brca1/2-unmutated HGSC modeling cells more tolerant to anti-tumor agents like carboplatin (p-value < 0.001), paclitaxel (p-value < 0.01), and olaparib (p-value < 0.01). To investigate biological processes that induce the aggressive and chemo-resistant phenotype, we performed gene set enrichment analysis among our HGSC models. The result showed that aberrant RAS/PI3K crosstalk induces cell cycle acceleration and downregulation of apoptosis in the HGSC cells, which backed up aggressive phenotype and chemo-resistance in our HGSC models. In conclusion, aberrant RAS/PI3K crosstalk transforms Brca1/2-unmutated HGSC modeling cells more aggressive and chemo-resistant, which was well recapitulated by our HGSC modeling organoids. The integration of our computational approach and experimental models will lead us to detect peculiar therapeutic targets against refractory HGSC. Citation Format: Tomohiro Tamura, Kenta Masuda, Shimpei Nagai, Keiyo Imaeda, Juntaro Yamasaki, Eiji Sugihara, Hiroyuki Nobusue, Yuji Otsuki, Rui Yamaguchi, Kazuhiro Sakurada, Wataru Yamagami, Hideyuki Saya, Osamu Nagano. Integration of human omics analysis and new syngeneic tumor organoid models reveals that aberrant RAS/PI3K crosstalk triggers refractoriness in high-grade serous carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1254.

Covariate-Dependent Clustering of Undirected Networks with Brain-Imaging Data

This article focuses on model-based clustering of subjects based on the shared relationships of subject-specific networks and covariates in scenarios when there are differences in the relationship between networks and covariates for different groups of subjects. It is also of interest to identify the network nodes significantly associated with each covariate in each cluster of subjects. To address these methodological questions, we propose a novel nonparametric Bayesian mixture modeling framework with an undirected network response and scalar predictors. The symmetric matrix coefficients corresponding to the scalar predictors of interest in each mixture component involve low-rankness and group sparsity within the low-rank structure. While the low-rank structure in the network coefficients adds parsimony and computational efficiency, the group sparsity within the low-rank structure enables drawing inference on network nodes and cells significantly associated with each scalar predictor. Being a principled Bayesian mixture modeling framework, our approach allows model-based identification of the number of clusters, offers clustering uncertainty in terms of the co-clustering matrix and presents precise characterization of uncertainty in identifying network nodes significantly related to a predictor in each cluster. Empirical results in various simulation scenarios illustrate substantial inferential gains of the proposed framework in comparison with competitors. Analysis of a real brain connectome dataset using the proposed method provides interesting insights into the brain regions of interest (ROIs) significantly related to creative achievement in each cluster of subjects. Supplementary material shows the convergence rate for the posterior predictive density of the proposed model, additional simulation examples with model misspecification, full conditional distributions to run the Markov chain Monte Carlo (MCMC) algorithm and also presents traceplots for various model parameters to demonstrate convergence of the MCMC algorithm.

Incidence of typhoid fever in Burkina Faso, Democratic Republic of the Congo, Ethiopia, Ghana, Madagascar, and Nigeria (the Severe Typhoid in Africa programme): a population-based study

Typhoid Fever remains a major cause of morbidity and mortality in low-income settings. The Severe Typhoid in Africa programme was designed to address regional gaps in typhoid burden data and identify populations eligible for interventions using novel typhoid conjugate vaccines. A hybrid design, hospital-based prospective surveillance with population-based health-care utilisation surveys, was implemented in six countries in sub-Saharan Africa. Patients presenting with fever (≥37·5°C axillary or ≥38·0°C tympanic) or reporting fever for three consecutive days within the previous 7 days were invited to participate. Typhoid fever was ascertained by culture of blood collected upon enrolment. Disease incidence at the population level was estimated using a Bayesian mixture model. 27 866 (33·8%) of 82 491 participants who met inclusion criteria were recruited. Blood cultures were performed for 27 544 (98·8%) of enrolled participants. Clinically significant organisms were detected in 2136 (7·7%) of these cultures, and 346 (16·2%) Salmonella enterica serovar Typhi were isolated. The overall adjusted incidence per 100 000 person-years of observation was highest in Kavuaya and Nkandu 1, Democratic Republic of the Congo (315, 95% credible interval 254-390). Overall, 46 (16·4%) of 280 tested isolates showed ciprofloxacin non-susceptibility. High disease incidence (ie, >100 per 100 000 person-years of observation) recorded in four countries, the prevalence of typhoid hospitalisations and complicated disease, and the threat of resistant typhoid strains strengthen the need for rapid dispatch and implementation of effective typhoid conjugate vaccines along with measures designed to improve clean water, sanitation, and hygiene practices. The Bill & Melinda Gates Foundation.

Rush to Charge, Dead to Drive: Application of Deadline Rush Model to Electric Vehicle User's Charging Behavior.

This work aims to estimate the portion of electric vehicle (EV) users who exhibit procrastination-like behavior, almost equivalent to an "empty" battery, before they decide to charge their vehicles. There is a human tendency to procrastinate when a deadline approaches. Human behavior in the presence of deadlines has been studied in different fields to evaluate individuals' performance or organizational efficiency and effectiveness. However, this phenomenon has not been investigated among EV users. This study explores users' procrastination-like behavior among 69 Rhode Island public charging stations' data representing 70,611 charging events. The Deadline Rush Model is incorporated to model frequent users' charging profiles. To conduct a robust estimation, the Bayesian Mixture Model is implemented. With the selection of an informative prior, the Bayesian Mixture Model estimated that almost one-third of frequent users procrastinate charging. The majority of procrastination-like users have small battery sizes. Although procrastination-like users need to charge when they arrive at a location, that might not necessarily be true for a plug-in hybrid; thus, systematically, they can clog the system for other users whose needs are more pressing. Understanding unique and unexplored charging behaviors among EV users is beneficial to EV infrastructure stakeholders in reducing the adoption threshold by providing a reliable and ubiquitous charging network. The findings identify a different kind of demand on the EV infrastructure than previously modeled and can directly influence future decision-making criteria in terms of planning to optimize to accommodate EV drivers with different charging behaviors.

A high‐dimensional multinomial logit model

SummaryThe number of parameters in a standard multinomial logit model increases linearly with the number of choice alternatives and number of explanatory variables. Because many modern applications involve large choice sets with categorical explanatory variables, which enter the model as large sets of binary dummies, the number of parameters in a multinomial logit model is often large. This paper proposes a new method for data‐driven two‐way parameter clustering over outcome categories and explanatory dummy categories in a multinomial logit model. A Bayesian Dirichlet process mixture model encourages parameters to cluster over the categories, which reduces the number of unique model parameters and provides interpretable clusters of categories. In an empirical application, we estimate the holiday preferences of 11 household types over 49 holiday destinations and identify a small number of household segments with different preferences across clusters of holiday destinations.

Identifying Brexit voting patterns in the British house of commons: an analysis based on Bayesian mixture models with flexible concomitant covariate effects

Abstract The results of some divisions related to Brexit held in the House of Commons are investigated. In particular, a new class of mixture models with concomitant covariates is developed to identify groups of members of parliament with similar voting behaviour. The methodological novelty lies in the flexibility introduced by the use of smooth functions to model the effect of concomitant covariates on the component weights of the mixture. Results show this approach allows to quantify the effect of the age of members of parliament, as well as preferences and competitiveness in the constituencies they represent, on their position towards Brexit.

A Bayesian finite mixture model approach to evaluate dichotomization method for correlated ELISA tests

In diagnostic accuracy studies, a commonly employed approach involves dichotomizing continuous data and subsequently analyzing them using a Bayesian latent class model (BLCM), often relying on binomial or multinomial distributions, rather than preserving their continuous nature. However, this procedure can inadvertently lead to less reliable outcomes due to the inherent loss of information when converting the original continuous measurements into binary values. Through comprehensive simulations, we demonstrated the limitations and disadvantages of dichotomizing continuous biomarkers from two correlated tests. Our findings highlighted notable disparities between the true values and the model estimates as a result of dichotomization. We discovered the crucial significance of selecting a reference test with high diagnostic accuracy in test evaluation in order to obtain reliable estimates of test accuracy and prevalences. Our study served as a call to action for veterinary researchers to exercise caution when utilizing dichotomization.

Bayesian Nonparametric Multivariate Mixture of Autoregressive Processes with Application to Brain Signals

One of neuroscience’s goals is to study the interactions between different brain regions during rest and while performing specific cognitive tasks. Multivariate Bayesian autoregressive decomposition (MBMARD) is proposed as an intuitive and novel Bayesian non-parametric model to represent high-dimensional signals as a low-dimensional mixture of univariate uncorrelated latent oscillations. Each latent oscillation captures a specific underlying oscillatory activity and, hence, is modeled as a unique second-order autoregressive process due to a compelling property, namely, that its spectral density’s shape is characterized by a unique frequency peak and bandwidth, parameterized by a location and a scale parameter. The posterior distributions of the latent oscillation parameters are computed using a Metropolis-within-Gibbs algorithm. One of the advantages of the MBMARD model is its higher robustness against misspecification, compared with standard models. The main scientific questions addressed by the MBMARD model relate to the effects of long-term alcohol abuse on memory. These effects were examined by analyzing the electroencephalogram records of alcoholic and non-alcoholic subjects performing a visual recognition experiment. The MBMARD model yielded novel and interesting findings, including the identification of subject-specific clusters of low- and high-frequency oscillations in different brain regions.

Variational Bayesian Gaussian mixture model for off‐grid DOA estimation

AbstractWireless signals are commonly subject to diverse and complex noise interference. The typical assumption of Gaussian white noise often oversimplifies the noise, resulting in reduced accuracy in estimating the direction of arrival (DOA), especially in complex scenarios. To tackle this issue, this paper introduces a new Bayesian model for off‐grid DOA estimation. This model utilizes Gaussian mixture model (GMM)‐based Dirichlet processes (DP) to characterize noise, allowing adaptive adjustments in the number of Gaussian mixture models. Leveraging the factor graph representation of the Bayesian model, a low‐complexity mixed messaging passing algorithm, employing generalized approximate message passing (GAMP) and mean field (MF), is proposed. Simulation results validate the efficacy of the proposed algorithm.

Incorporating Prior Data in Quantitative Benefit-Risk Assessments: Case Study of a Bayesian Method.

Multiple criteria decision analysis (MCDA) and stochastic multi-criteria acceptability analysis (SMAA) in their current implementation cannot incorporate prior or external information on benefits and risks. We demonstrate how to incorporate prior data using a Bayesian mixture model approach while conducting quantitative benefit-risk assessments (qBRA) for medical products. We implemented MCDA and SMAA in a Bayesian framework. To incorporate information from a prior study, we use mixture priors on each benefit and risk attribute that mixes information from a previous study with a vague prior distribution. The degree of borrowing is varied using a mixing proportion parameter. A demonstration case study for qBRA using the supplementary New Drug Application (sNDA) filing for Rivaroxaban for the indication of reduction in the risk of major thrombotic vascular events in patients with peripheral artery disease (PAD) was used to illustrate the method. Net utility scores, obtained from the randomized controlled trial data to support the sNDA, from the MCDA for Rivaraxoban and comparator were 0.48 and 0.56, respectively, with Rivaroxaban being the preferred alternative only 33% of the time. We show that with only 30% borrowing from a previous RCT, the MCDA and SMAA results are favorable for Rivaroxaban, accounting for the seemingly aberrant results on all-cause death in the trial data used to support the sNDA. Our method to formally incorporate prior data in MCDA and SMAA is easy to use and interpret. Software in the form of an RShiny App is available here: https://sai-dharmarajan.shinyapps.io/BayesianMCDA_SMAA/ .

Bayesian response adaptive randomization design with a composite endpoint of mortality and morbidity.

Allocating patients to treatment arms during a trial based on the observed responses accumulated up to the decision point, and sequential adaptation of this allocation, could minimize the expected number of failures or maximize total benefits to patients. In this study, we developed a Bayesian response-adaptive randomization (RAR) design targeting the endpoint of organ support-free days (OSFD) for patients admitted to the intensive care units. The OSFD is a mixture of mortality and morbidity assessed by the number of days of free of organ support within a predetermined post-randomization time-window. In the past, researchers treated OSFD as an ordinal outcome variable where the lowest category is death. We propose a novel RAR design for a composite endpoint of mortality and morbidity, for example, OSFD, by using a Bayesian mixture model with a Markov chain Monte Carlo sampling to estimate the posterior probability distribution of OSFD and determine treatment allocation ratios at each interim. Simulations were conducted to compare the performance of our proposed design under various randomization rules and different alpha spending functions. The results show that our RAR design using Bayesian inference allocated more patients to the better performing arm(s) compared to other existing adaptive rules while assuring adequate power and type I error rate control across a range of plausible clinical scenarios.

Ant backbone phylogeny resolved by modelling compositional heterogeneity among sites in genomic data

Ants are the most ubiquitous and ecologically dominant arthropods on Earth, and understanding their phylogeny is crucial for deciphering their character evolution, species diversification, and biogeography. Although recent genomic data have shown promise in clarifying intrafamilial relationships across the tree of ants, inconsistencies between molecular datasets have also emerged. Here I re-examine the most comprehensive published Sanger-sequencing and genome-scale datasets of ants using model comparison methods that model among-site compositional heterogeneity to understand the sources of conflict in phylogenetic studies. My results under the best-fitting model, selected on the basis of Bayesian cross-validation and posterior predictive model checking, identify contentious nodes in ant phylogeny whose resolution is modelling-dependent. I show that the Bayesian infinite mixture CAT model outperforms empirical finite mixture models (C20, C40 and C60) and that, under the best-fitting CAT-GTR + G4 model, the enigmatic Martialis heureka is sister to all ants except Leptanillinae, rejecting the more popular hypothesis supported under worse-fitting models, that place it as sister to Leptanillinae. These analyses resolve a lasting controversy in ant phylogeny and highlight the significance of model comparison and adequate modelling of among-site compositional heterogeneity in reconstructing the deep phylogeny of insects.

A Bayesian fine-mapping model using a continuous global-local shrinkage prior with applications in prostate cancer analysis

The aim of fine mapping is to identify genetic variants causally contributing to complex traits or diseases. Existing fine-mapping methods employ Bayesian discrete mixture priors and depend on a pre-specified maximum number of causal variants, which may lead to sub-optimal solutions. In this work, we propose a Bayesian fine-mapping method called h2-D2, utilizing a continuous global-local shrinkage prior. We also present an approach to define credible sets of causal variants in continuous prior settings. Simulation studies demonstrate that h2-D2 outperforms current state-of-the-art fine-mapping methods such as SuSiE and FINEMAP in accurately identifying causal variants and estimating their effect sizes. We further applied h2-D2 to prostate cancer analysis and discovered some previously unknown causal variants. In addition, we inferred 369 target genes associated with the detected causal variants and several pathways that were significantly over-represented by these genes, shedding light on their potential roles in prostate cancer development and progression.

A Bayesian Gaussian Mixture Model for Probabilistic Modeling of Car-Following Behaviors

A Bayesian Gaussian Mixture Model for Probabilistic Modeling of Car-Following Behaviors

Documenting and modeling the acoustic variability of intervocalic alveolar taps in conversational Peninsular Spanish.

This study constitutes an investigation into the acoustic variability of intervocalic alveolar taps in a corpus of spontaneous speech from Madrid, Spain. Substantial variability was documented in this segment, with highly reduced variants constituting roughly half of all tokens during spectrographic inspection. In addition to qualitative documentation, the intensity difference between the tap and surrounding vowels was measured. Changes in this intensity difference were statistically modeled using Bayesian finite mixture models containing lexical and phonetic predictors. Model comparisons indicate predictive performance is improved when we assume two latent categories, interpreted as two pronunciation variants for the Spanish tap. In interpreting the model, predictors were more often related to categorical changes in which pronunciation variant was produced than to gradient intensity changes within each tap type. Variability in tap production was found according to lexical frequency, speech rate, and phonetic environment. These results underscore the importance of evaluating model fit to the data as well as what researchers modeling phonetic variability can gain in moving past linear models when they do not adequately fit the observed data.

Estimating the prevalence of diabetic retinopathy in electronic health records with massive missing labels

ObjectiveThe paper aims to address the problem of massive unlabeled patients in electronic health records (EHR) who potentially have undiagnosed diabetic retinopathy (DR). It is desired to estimate the actual DR prevalence in EHR with 96 % missing labels. Materials and methodsThe Cerner Health Facts data are used in the study, with 3749 labeled DR patients and 97,876 unlabeled diabetic patients. This extensive dataset spans the demographics of the United States over the past two decades. We implemented state-of-art positive-unlabeled learning methods, including ensemble-based support vector machine, ensemble-based random forest, and Bayesian finite mixture modeling. ResultsThe estimated DR prevalence in the population represented by Cerner EHR is approximately 25 % and the classification techniques generally achieve an AUC of around 87 %. As a by-product, a predictive inference on the risk of DR based on a patient's personalized medical information is derived. DiscussionMissing labels is a common issue for EHR data quality. Ignoring these missing labels can lead to biased results in the analyses of EHR data. The problem is especially severe in the context of DR. It is thus important to use machine learning or statistical tools to identify the unlabeled patients. The tool in this paper helps both data analysts and clinicians in their practices.