The two-sided markets, such as ride-sharing companies, often involve a group of subjects who are making sequential decisions across time and/or location. With the rapid development of smart phones and internet of things, they have substantially transformed the transportation landscape of human beings. In this paper we consider large-scale fleet management in ride-sharing companies that involve multiple units in different areas receiving sequences of products (or treatments) over time. Major technical challenges, such as policy evaluation, arise in those studies because: (i) spatial and temporal proximities induce interference between locations and times, and (ii) the large number of locations results in the curse of dimensionality. To address both challenges simultaneously, we introduce a multiagent reinforcement learning (MARL) framework for carrying policy evaluation in these studies. We propose novel estimators for mean outcomes under different products that are consistent despite the high dimensionality of state-action space. The proposed estimator works favorably in simulation experiments. We further illustrate our method using a real dataset obtained from a two-sided marketplace company to evaluate the effects of applying different subsidizing policies. A Python implementation of our proposed method is available in the Supplementary Material and also at https://github.com/RunzheStat/CausalMARL.

Characterizing the cumulative burden of COVID-19 by race/ethnicity is of the utmost importance for public health researchers and policy makers in order to design effective mitigation measures. This analysis is hampered, however, by surveillance case data with substantial missingness in race and ethnicity covariates. Worse yet, this missingness likely depends on the values of these missing covariates; that is, they are not-missing-at-random (NMAR). We propose a Bayesian parametric model that leverages joint information on spatial variation in the disease and covariate missingness processes and can accommodate both MAR and NMAR missingness. We show that the model is locally identifiable when the spatial distribution of the population covariates is known and observed cases can be associated with a spatial unit of observation. We also use a simulation study to investigate the model’s finite-sample performance. We compare our model’s performance on NMAR data against complete-case analysis and multiple imputation (MI), both of which are commonly used by public health researchers when confronted with missing categorical covariates. Finally, we model spatial variation in cumulative COVID-19 incidence in Wayne County, Michigan, using data from the Michigan Department of Health and Human Services. The analysis suggests that population relative risk estimates by race during the early part of the COVID-19 pandemic in Michigan were understated for non-white residents, compared to white residents, when cases missing race were dropped or had these values imputed using MI.

A central goal in experimental high energy physics is to detect new physics signals that are not explained by known physics. In this paper we aim to search for new signals that appear as deviations from known Standard Model physics in high-dimensional particle physics data. To do this, we determine whether there is any statistically significant difference between the distribution of Standard Model background samples and the distribution of the experimental observations which are a mixture of the background and a potential new signal. Traditionally, one also assumes access to a sample from a model for the hypothesized signal distribution. Here we instead investigate a model-independent method that does not make any assumptions about the signal and uses a semisupervised classifier to detect the presence of the signal in the experimental data. We construct three test statistics using the classifier: an estimated likelihood ratio test (LRT) statistic, a test based on the area under the ROC curve (AUC), and a test based on the misclassification error (MCE). Additionally, we propose a method for estimating the signal strength parameter and explore active subspace methods to interpret the proposed semisupervised classifier in order to understand the properties of the detected signal. We also propose a score test statistic that can be used in the model-dependent setting. We investigate the performance of the methods on a simulated data set related to the search for the Higgs boson at the Large Hadron Collider at CERN. We demonstrate that the semisupervised tests have power competitive with the classical supervised methods for a well-specified signal but much higher power for an unexpected signal which might be entirely missed by the supervised tests.

Recent statistical methods fitted on large-scale GPS data can provide accurate estimations of the expected travel time between two points. However, little is known about the distribution of travel time, which is key to decision-making across a number of logistic problems. With sufficient data single road-segment travel time can be well approximated. The challenge lies in understanding how to aggregate such information over a route to arrive at the route-distribution of travel time. We develop a novel statistical approach to this problem. We show that, under general conditions and without assuming a distribution of speed, travel time divided by route distance follows a Gaussian distribution with route-invariant population mean and variance. We develop efficient inference methods for these parameters and propose asymptotically tight population prediction intervals for travel time. Using traffic flow information, we further develop a trip-specific Gaussian-based predictive distribution, resulting in tight prediction intervals for short and long trips. Our methods, implemented in an R-package,¹ are illustrated in a real-world case study using mobile GPS data, showing that our trip-specific and population intervals both achieve the 95% theoretical coverage levels. Compared to alternative approaches, our trip-specific predictive distribution achieves: (a) the theoretical coverage at every level of significance, (b) tighter prediction intervals, (c) less predictive bias, and (d) more efficient estimation and prediction procedures. This makes our approach promising for low-latency, large-scale transportation applications.

¹Available at https://github.com/melmasri/traveltimeCLT.

Biometric traits, such as fingerprints, facial images, and teeth impressions, are often used in forensic analysis to identify crime suspects. Matching such biometric traits is not perfect, and recent reports have indicated the need for quantifiable measures of error rates for (these) possible matches. Often, comparisons between two sets of a trait are scored with a higher score indicating a higher likelihood that the sets are a match. Adjustment of the cutoff for which a match is declared yields a trade-off between false positive and false negative decisions that can be represented by an ROC curve. In this paper we study modeling of such ROC curves conditional on covariates, for example, demographic information about source subjects, quality properties of the underlying biometric measurements, or characteristics of forensic examiners; quantifying how error rates vary in dependence of such covariates is often considerably more meaningful in biometrics and forensics than the “raw” error rates based on the pooled data. We herein develop a framework for estimating covariate-specific ROC curves that integrates robustness, heteroscedasticity, and stochastic ordering. The latter is of specific relevance in the given application since biometric recognition systems are typically calibrated to assign higher scores to matching pairs than to nonmatching pairs. The proposed methodology is demonstrated on accuracy of face recognition and fingerprint matching and also has potential in other domains of application like medical diagnostics.

In a pilot program during the 2016–17 admissions cycle, the University of California, Berkeley invited many applicants for freshman admission to submit letters of recommendation. This proved controversial within the university, with concerns that this change would further disadvantage applicants from disadvantaged groups. To inform this debate, we use this pilot as the basis for an observational study of the impact of submitting letters of recommendation on subsequent admission, with the goal of estimating how impacts vary across predefined subgroups. Understanding this variation is challenging in an observational setting because estimated impacts reflect both actual treatment effect variation and differences in covariate balance across groups. To address this, we develop balancing weights that directly optimize for “local balance” within subgroups while maintaining global covariate balance between treated and control units. Applying this approach to the UC Berkeley pilot study yields excellent local and global balance, unlike more traditional weighting methods, which fail to balance covariates within subgroups. We find that the impact of letters of recommendation increases with applicant strength. However, we find little average difference for applicants from disadvantaged groups, although this result is more mixed. In the end we conclude that soliciting letters of recommendation from a broader pool of applicants would not meaningfully change the composition of admitted undergraduates.

The majority of Americans fail to achieve recommended levels of physical activity, which leads to numerous preventable health problems, such as diabetes, hypertension, and heart diseases. This has generated substantial interest in monitoring human activity to gear interventions toward environmental features that may relate to higher physical activity. Wearable devices, such as wrist-worn sensors that monitor gross motor activity (actigraph units) continuously record the activity levels of a subject, producing massive amounts of high-resolution measurements. Analyzing actigraph data needs to account for spatial and temporal information on trajectories or paths traversed by subjects wearing such devices. Inferential objectives include estimating a subject’s physical activity levels along a given trajectory, identifying trajectories that are more likely to produce higher levels of physical activity for a given subject, and predicting expected levels of physical activity in any proposed new trajectory for a given set of health attributes. Here, we devise a Bayesian hierarchical modeling framework for spatial-temporal actigraphy data to deliver fully model-based inference on trajectories while accounting for subject-level health attributes and spatial-temporal dependencies. We undertake a comprehensive analysis of an original dataset from the Physical Activity through Sustainable Transport Approaches in Los Angeles (PASTA-LA) study to ascertain spatial zones and trajectories exhibiting significantly higher levels of physical activity while accounting for various sources of heterogeneity.

While vaccines are crucial to end the COVID-19 pandemic, public confidence in vaccine safety has always been vulnerable. Many statistical methods have been applied to VAERS (Vaccine Adverse Event Reporting System) database to study the safety of COVID-19 vaccines. However, none of these methods considered the adverse event (AE) ontology. AEs are naturally related; for example, events of retching, dysphagia, and reflux are all related to an abnormal digestive system. Explicitly bringing AE relationships into the model can aid in the detection of true AE signals amid the noise while reducing false positives. We propose a Bayesian graph-assisted signal selection (BGrass) model to simultaneously estimate all AEs while incorporating the network of dependence between AEs. Under a fully Bayesian inference framework, we also propose a negative control approach to mitigate the reporting bias and an enrichment approach to detecting AE groups of concern. For posterior computation we construct an equivalent model representation and develop an efficient Gibbs sampler. We evaluate the performance of BGrass via extensive simulations. To study the safety of COVID-19 vaccines, we apply BGrass to analyze approximately one million VAERS reports (01/01/2016–12/24/2021) involving more than 800 AEs. In particular, we found that blood clots (including deep vein thrombosis, thrombosis, and pulmonary embolism) are more likely to be reported after COVID-19 vaccination, compared to influenza vaccines. They are also reported more often for Johnson & Johnson–Janssen vaccine, compared to mRNA-based COVID-19 vaccines. A user-friendly $\mathtt{R}$ package $\mathtt{BGrass}$ that implements the proposed methods to assess vaccine safety is included in the Supplementary Material and is publicly available at https://github.com/BangyaoZhao/BGrass.

Coronavirus case-count data has influenced government policies and drives most epidemiological forecasts. Limited testing is cited as the key driver behind minimal information on the COVID-19 pandemic. While expanded testing is laudable, measurement error and selection bias are the two greatest problems limiting our understanding of the COVID-19 pandemic; neither can be fully addressed by increased testing capacity. In this paper we demonstrate their impact on estimation of point prevalence and the effective reproduction number. We show that estimates, based on the millions of molecular tests in the U.S., have the same mean square error as a small simple random sample. To address this, a procedure is presented that combines case-count data and random samples over time to estimate selection propensities based on key covariate information. We then combine these selection propensities with epidemiological forecast models to construct a doubly robust estimation method that accounts for both measurement-error and selection bias. This method is then applied to estimate Indiana’s active infection prevalence using case-count, hospitalization, and death data with demographic information, a statewide random molecular sample collected from April 25–29, 2020, and Delphi’s COVID-19 Trends and Impact Survey. We end with a series of recommendations based on the proposed methodology.

In The Cancer Genome Atlas (TCGA) data set, there are many interesting nonlinear dependencies between pairs of genes that reveal important relationships and subtypes of cancer. Such genomic data analysis requires a rapid, powerful, and interpretable detection process, especially in a high-dimensional environment. We study the nonlinear patterns among the expression of pairs of genes from TCGA using a powerful tool called binary expansion testing. We find many nonlinear patterns, some of which are driven by known cancer subtypes, some of which are novel.

Motivated by emerging applications in ecology, microbiology, and neuroscience, this paper studies high-dimensional regression with two-way structured data. To estimate the high-dimensional coefficient vector, we propose the generalized matrix decomposition regression (GMDR) to efficiently leverage auxiliary information on row and column structures. GMDR extends the principal component regression (PCR) to two-way structured data, but unlike PCR, GMDR selects the components that are most predictive of the outcome, leading to more accurate prediction. For inference on regression coefficients of individual variables, we propose the generalized matrix decomposition inference (GMDI), a general high-dimensional inferential framework for a large family of estimators that include the proposed GMDR estimator. GMDI provides more flexibility for incorporating relevant auxiliary row and column structures. As a result, GMDI does not require the true regression coefficients to be sparse but constrains the coordinate system representing the regression coefficients according to the column structure. GMDI also allows dependent and heteroscedastic observations. We study the theoretical properties of GMDI in terms of both the type-I error rate and power and demonstrate the effectiveness of GMDR and GMDI in simulation studies and an application to human microbiome data.

The limited representation of minorities and disadvantaged populations in large-scale clinical and genomics research poses a significant barrier to translating precision medicine research into practice. Prediction models are likely to underperform in underrepresented populations due to heterogeneity across populations, thereby exacerbating known health disparities. To address this issue, we propose FETA, a two-way data integration method that leverages a federated transfer learning approach to integrate heterogeneous data from diverse populations and multiple healthcare institutions, with a focus on a target population of interest having limited sample sizes. We show that FETA achieves performance comparable to the pooled analysis, where individual-level data is shared across institutions, with only a small number of communications across participating sites. Our theoretical analysis and simulation study demonstrate how FETA’s estimation accuracy is influenced by communication budgets, privacy restrictions, and heterogeneity across populations. We apply FETA to multisite data from the electronic Medical Records and Genomics (eMERGE) Network to construct genetic risk prediction models for extreme obesity. Compared to models trained using target data only, source data only, and all data without accounting for population-level differences, FETA shows superior predictive performance. FETA has the potential to improve estimation and prediction accuracy in underrepresented populations and reduce the gap in model performance across populations.

In clinical practice medications are often interchanged in treatment protocols when a patient negatively reacts to their first line of therapy. Although switching between medications is common, clinicians often lack structured guidance when choosing the initial dose and frequency of a new medication, given the former with respect to risk of adverse events. In this paper we propose to establish this dose toxo-equivalence relationship using published clinical trial results with one or both drugs of interest via a Bayesian meta-analysis model that accounts for both within- and between-study variances. With the posterior parameter samples from this model, we compute median and $95\mathrm{\%}$ credible intervals for equivalent dose pairs of the two drugs that are predicted to produce equal rates of an adverse outcome, relying solely on study-level information. Via extensive simulations, we show that this approach approximates well the true dose toxo-equivalence relationship, considering different study designs, levels of between-study variance, and the inclusion/exclusion of nonconfounder/nonmodifier subject-level covariates in addition to study-level covariates. We compare the performance of this study-level meta-analysis estimate to the equivalent individual patient data meta-analysis model and find comparable bias and minimal efficiency loss in the study-level coefficients used in the dose toxo-equivalence relationship. Finally, we present the findings of our dose toxo-equivalence model applied to two chemotherapy drugs, based on data from 169 published clinical trials.

Volumetric imaging features are used in cancer research to determine the size and the composition of a tumor and have been shown to be prognostic of overall survival. In this paper we focus on the analysis of tumor component proportions of brain cancer patients collected through The Cancer Genome Atlas (TCGA) project. Our main goal is to identify pathways and corresponding genes that can explain the heterogeneity of the composition of a brain tumor. In particular, we focus on the glioblastoma multiform (GBM), as it is the most common malignant brain neoplasm, accounting for 23% of all primary brain tumors for which it still has very poor prognosis. We propose a Bayesian hierarchical model for variable selection with a group structure in the context of correlated multivariate compositional response variables. More specifically, we model the proportions of the tumor components within the tumor using a Dirichlet model by allowing for straightforward incorporation of available high-dimensional covariate information within a log-linear regression framework. We impose prior distributions that account for the overlapping structure between groups of covariates. Simulations and application to GBM disease show the importance of our approach. We have identified associations between tumor component volume-based features and several important pathways and genes. Some of these genes have previously been shown to be prognostic indicators of overall survival time in GBM.

Numerous adverse effects (e.g., depression) have been reported for combination antiretroviral therapy (cART) despite its remarkable success in viral suppression in people with HIV (PWH). To improve long-term health outcomes for PWH, there is an urgent need to design personalized optimal cART with the lowest risk of comorbidity in the emerging field of precision medicine for HIV. Large-scale HIV studies offer researchers unprecedented opportunities to optimize personalized cART in a data-driven manner. However, the large number of possible drug combinations for cART makes the estimation of cART effects a high-dimensional combinatorial problem, imposing challenges in both statistical inference and decision-making. We develop a two-step Bayesian decision framework for optimizing sequential cART assignments. In the first step, we propose a dynamic model for individuals’ longitudinal observations using a multivariate Gaussian process. In the second step, we build a probabilistic generative model for cART assignments and design an uncertainty-penalized policy optimization using the uncertainty quantification from the first step. Applying the proposed method to a dataset from the Women’s Interagency HIV Study, we demonstrate its clinical utility in assisting physicians to make effective treatment decisions, serving the purpose of both viral suppression and comorbidity risk reduction.

Low-cost air pollution sensors, offering hyperlocal characterization of pollutant concentrations, are becoming increasingly prevalent in environmental and public health research. However, low-cost air pollution data can be noisy, biased by environmental conditions, and usually need to be field-calibrated by collocating low-cost sensors with reference-grade instruments. We show, theoretically and empirically, that the common procedure of regression-based calibration, using collocated data, systematically underestimates high air pollution concentrations, which are critical to diagnose from a health perspective. Current calibration practices also often fail to utilize the spatial correlation in pollutant concentrations. We propose a novel spatial filtering approach to collocation-based calibration of low-cost networks that mitigates the underestimation issue by using an inverse regression. The inverse regression also allows for incorporating spatial correlations by a second-stage model for the true pollutant concentrations using a conditional Gaussian process. Our approach works with one or more collocated sites in the network and is dynamic, leveraging spatial correlation with the latest available reference data. Through extensive simulations, we demonstrate how the spatial filtering substantially improves estimation of pollutant concentrations and measures peak concentrations with greater accuracy. We apply the methodology for calibration of a low-cost PM${}_{2.5}$ network in Baltimore, Maryland, and diagnose air pollution peaks that are missed by the regression-calibration.

Chimney fires constitute one of the most commonly occurring fire types. Precise prediction and prompt prevention are crucial in reducing the harm they cause. In this paper we develop a combined machine learning and statistical modelling process to predict fire risk. First, we use random forests and permutation importance techniques to identify the most informative explanatory variables. Second, we design a Poisson point process model and employ logistic regression estimation to estimate the parameters. Moreover, we validate the Poisson model assumption using second-order summary statistics and residuals. We implement the modelling process on data collected by the Twente Fire Brigade and obtain plausible predictions. Compared to similar studies, our approach has two advantages: (i) with random forests, we can select explanatory variables nonparametrically considering variable dependence; (ii) using logistic regression estimation, we can fit our statistical model efficiently by tuning it to focus on regions and times that are salient for fire risk.

We consider the challenges that arise when fitting ecological individual heterogeneity models to “large” data sets. In particular, we focus on (continuous-valued) random effect models commonly used to describe individual heterogeneity present in ecological populations within the context of capture–recapture data, although the approach is more widely applicable to more general latent variable models. Within such models the associated likelihood is expressible only as an analytically intractable integral. Common techniques for fitting such models to data include, for example, the use of numerical approximations for the integral or a Bayesian data augmentation approach. However, as the size of the data set increases (i.e., the number of individuals increases), these computational tools may become computationally infeasible. We present an efficient Bayesian model-fitting approach, whereby we initially sample from the posterior distribution of a smaller subsample of the data, before correcting this sample to obtain estimates of the posterior distribution of the full data set using an importance sampling approach. We consider several practical issues, including the subsampling mechanism, computational efficiencies (including the ability to parallelise the algorithm) and combining subsampling estimates using multiple subsampled data sets. We initially demonstrate the feasibility (and accuracy) of the approach via simulated data before considering a challenging real data set of approximately 30,000 guillemots and, using the proposed algorithm, obtain posterior estimates of the model parameters in substantially reduced computational time, compared to the standard Bayesian model-fitting approach.

Land use/land cover mapping is usually performed by classifying satellite imagery (e.g., Landsat, Sentinel) for the whole survey region using classification algorithms implemented with training data. Subsequently, probabilistic samples are usually implemented with the main purpose of assessing the accuracy of these maps by comparing the map class and the ground condition determined for the sampled units. The main proposal of this paper is to directly exploit these probabilistic samples to estimate the land use/land cover class at any location of the survey region in a design-based framework by the well-known nearest-neighbour interpolator. For the first time, the design-based consistency of nearest-neighbour maps (i.e., categorical variables) is theoretically proven and a pseudo-population bootstrap estimator of their precision is proposed and discussed. These nearest-neighbour maps provide the ability to place mapping within a rigorous design-based inference framework, in contrast to most traditional mapping approaches which often are implemented with no inferential basis or by necessity (due to lack of a probabilistic sample) model-based inference. A simulation study is performed on an estimated land use map in Southern Tuscany (Italy)—taken as the true map—to check the finite-sample performance of the proposal as well as the matching of the area coverage estimates arising from the map with those achieved by traditional estimators. The Italian land use map arising from the IUTI surveys and the U.S. land cover map arising from the LCMAP program are considered as case studies.

Spatially aggregated disease-count data relating to a set of nonoverlapping areal units are often used to make inference on population-level disease risk. This includes the identification of risk boundaries, which are locations where there is a sizeable change in risk between geographically neighbouring areal units. Existing studies provide spatially discrete inference on the areal unit footprint, which forces the boundaries to coincide with the entire geographical border between neighbouring units. This paper is the first to relax these assumptions by estimating disease risk and the locations of risk boundaries on a grid of square pixels covering the study region that can be made arbitrarily small to approximate a spatially continuous surface. We propose a two-stage approach that first fits a Bayesian spatiotemporal realignment model to estimate disease risk at the grid level and then identifies boundaries in this surface using edge detection algorithms from computer vision. This novel methodological fusion is motivated by a new study of respiratory hospitalisation risk in Glasgow, Scotland, between 2008 and 2017, and we identify numerous risk boundaries across the city.

Modeling and forecasting earthquakes is challenging due to the complex interplay and clustering of main-shocks and aftershocks. The epidemic-type aftershock sequence (ETAS) model represents the conditional intensity of earthquakes as the superposition of a background and aftershock rate which allows for the declustering of the earthquakes. Its success has led to the development of numerous versions of the ETAS model. Among these extensions is the renewal ETAS (RETAS) model, which has shown promising potential. The RETAS model endows the main-shock arrival process with a renewal process, which serves as an alternative to the homogeneous Poisson process. Model fitting is performed using likelihood-based estimation by directly optimizing the exact likelihood. However, inferring the branching structure from the fitted RETAS model remains a challenging task since the declustering algorithm that is currently available for the ETAS model is not directly applicable. Therefore, this article develops an iterative algorithm to calculate the smoothed main- and aftershock probabilities, conditional on all available information contained in the catalog. Consequently, an estimate of the background spatial intensity function and model parameters can be obtained using an iterative semiparametric procedure with the smoothing parameters selected using information criteria. The methods proposed herein are illustrated on simulated data and a New Zealand earthquake catalog.

The Internet of Things (IoT) system generates massive high-speed temporally correlated streaming data and is often connected with online inference tasks under computational or energy constraints. Online analysis of these streaming time series data often faces a trade-off between statistical efficiency and computational cost. One important approach to balance this trade-off is sampling, where only a small portion of the sample is selected for the model fitting and update. Motivated by the demands of dynamic relationship analysis of IoT system, we study the data-dependent sample selection and online inference problem for a multidimensional streaming time series, aiming to provide low-cost real-time analysis of high-speed power grid electricity consumption data. Inspired by D-optimality criterion in design of experiments, we propose a class of online data reduction methods that achieve an optimal sampling criterion and improve the computational efficiency of the online analysis. We show that the optimal solution amounts to a strategy that is a mixture of Bernoulli sampling and leverage score sampling. The leverage score sampling involves auxiliary estimations that have a computational advantage over recursive least squares updates. Theoretical properties of the auxiliary estimations involved are also discussed. When applied to European power grid consumption data, the proposed leverage score based sampling methods outperform the benchmark sampling method in online estimation and prediction. The general applicability of the sampling-assisted online estimation method is assessed via simulation studies.

Physical activity (PA) is significantly associated with many health outcomes. The wide usage of wearable accelerometer-based activity trackers in recent years has provided a unique opportunity for in-depth research on PA and its relations with health outcomes and interventions. Past analysis of activity tracker data relies heavily on aggregating minute-level PA records into day-level summary statistics in which important information of PA temporal/diurnal patterns is lost. In this paper we propose a novel functional data analysis approach based on Riemann manifolds for modeling PA and its longitudinal changes. We model smoothed minute-level PA of a day as one-dimensional Riemann manifolds and longitudinal changes in PA in different visits as deformations between manifolds. The variability in changes of PA among a cohort of subjects is characterized via variability in the deformation. Functional principal component analysis is further adopted to model the deformations, and PC scores are used as a proxy in modeling the relation between changes in PA and health outcomes and/or interventions. We conduct comprehensive analyses on data from two clinical trials: Reach for Health (RfH) and Metabolism, Exercise and Nutrition at UCSD (MENU), focusing on the effect of interventions on longitudinal changes in PA patterns and how different modes of changes in PA influence weight loss, respectively. The proposed approach reveals unique modes of changes, including overall enhanced PA, boosted morning PA, and shifts of active hours specific to each study cohort. The results bring new insights into the study of longitudinal changes in PA and health and have the potential to facilitate designing of effective health interventions and guidelines.

A key task in the emerging field of materials informatics is to use machine learning to predict a material’s properties and functions. A fast and accurate predictive model allows researchers to more efficiently identify or construct a material with desirable properties. As in many fields, deep learning is one of the state-of-the art approaches, but fully training a deep learning model is not always feasible in materials informatics due to limitations on data availability, computational resources, and time. Accordingly, there is a critical need in the application of deep learning to materials informatics problems to develop efficient transfer learning algorithms. The Bayesian framework is natural for transfer learning because the model trained from the source data can be encoded in the prior distribution for the target task of interest. However, the Bayesian perspective on transfer learning is relatively unaccounted for in the literature and is complicated for deep learning because the parameter space is large and the interpretations of individual parameters are unclear. Therefore, rather than subjective prior distributions for individual parameters, we propose a new Bayesian transfer learning approach based on the penalized complexity prior on the Kullback–Leibler divergence between the predictive models of the source and target tasks. We show via simulations that the proposed method outperforms other transfer learning methods across a variety of settings. The proposed method is applied to predict the properties of a molecular crystal, based on its structural properties, and we show improved precision for estimating the band gap of a material compared to state-of-the-art methods currently used in materials science.

Recent evidence highlights the usefulness of DNA methylation (DNAm) biomarkers as surrogates for exposure to risk factors for noncommunicable diseases in epidemiological studies and randomized trials. DNAm variability has been demonstrated to be tightly related to lifestyle behavior and exposure to environmental risk factors, ultimately providing an unbiased proxy of an individual state of health. At present, the creation of DNAm surrogates relies on univariate penalized regression models, with elastic-net regularizer being the gold standard when accomplishing the task. Nonetheless, more advanced modeling procedures are required in the presence of multivariate outcomes with a structured dependence pattern among the study samples. In this work we propose a general framework for mixed-effects multitask learning in presence of high-dimensional predictors to develop a multivariate DNAm biomarker from a multicenter study. A penalized estimation scheme, based on an expectation-maximization algorithm, is devised in which any penalty criteria for fixed-effects models can be conveniently incorporated in the fitting process. We apply the proposed methodology to create novel DNAm surrogate biomarkers for multiple correlated risk factors for cardiovascular diseases and comorbidities. We show that the proposed approach, modeling multiple outcomes together, outperforms state-of-the-art alternatives both in predictive power and biomolecular interpretation of the results.

Motivated by a study of United Nations voting behaviors, we introduce a regression model for a series of networks that are correlated over time. Our model is a dynamic extension of the additive and multiplicative effects network model (AMEN) of Hoff (Statist. Sci. 36 (2021) 34–50). In addition to incorporating a temporal structure, the model accommodates two types of missing data and thus allows the size of the network to vary over time. We demonstrate via simulations the necessity of various components of the model. We apply the model to the United Nations General Assembly voting data from 1983 to 2014 (In Routledge Handbook of International Organization (2013) Routledge) to answer interesting research questions regarding international voting behaviors. In addition to finding important factors that could explain the voting behaviors, the model-estimated additive effects, multiplicative effects, and their movements reveal meaningful foreign policy positions and alliances of various countries.

Random sampling of graph partitions under constraints has become a popular tool for evaluating legislative redistricting plans. Analysts detect partisan gerrymandering by comparing a proposed redistricting plan with an ensemble of sampled alternative plans. For successful application sampling methods must scale to maps with a moderate or large number of districts, incorporate realistic legal constraints, and accurately and efficiently sample from a selected target distribution. Unfortunately, most existing methods struggle in at least one of these areas. We present a new sequential Monte Carlo (SMC) algorithm that generates a sample of redistricting plans converging to a realistic target distribution. Because it draws many plans in parallel, the SMC algorithm can efficiently explore the relevant space of redistricting plans better than the existing Markov chain Monte Carlo (MCMC) algorithms that generate plans sequentially. Our algorithm can simultaneously incorporate several constraints commonly imposed in real-world redistricting problems, including equal population, compactness, and preservation of administrative boundaries. We validate the accuracy of the proposed algorithm by using a small map where all redistricting plans can be enumerated. We then apply the SMC algorithm to evaluate the partisan implications of several maps submitted by relevant parties in a recent high-profile redistricting case in the State of Pennsylvania. We find that the proposed algorithm converges faster and with fewer samples than a comparable MCMC algorithm. Open-source software is available for implementing the proposed methodology.

The real-world performance of vaccines against COVID-19 infections is critically important to counter the pandemics. We propose a varying coefficient stochastic epidemic model to estimate the vaccine protection rates based on the publicly available epidemiological and vaccination data. To tackle the challenges posed by the unobserved state variables, we develop a multistep decentralized estimation procedure that uses different data segments to estimate different parameters. A B-spline structure is used to approximate the underlying infection rates and to facilitate model simulation in obtaining an objective function between the imputed and the simulation-based estimates of the latent state variables, leading to simulation-based estimation of the diagnosis rate using data in the prevaccine period and the vaccine effect parameters using data in the postvaccine periods. The time-varying infection, recovery and death rates are estimated by kernel regressions. We apply the proposed method to analyze the data in ten countries which collectively used eight vaccines. The analysis reveals that the average protection rate of the full vaccination was at least 22% higher than that of the partial vaccination and was largely above the WHO recognized level of 50% before November 20, 2021, including the Delta variant dominated period. The protection rates for the booster vaccine in the Omicron period were also provided.

The question addressed here is whether, using Hawkes models. the distribution of SARS-CoV-2 (Covid-19) transmission times can be estimated accurately with only case-count data. We fit Hawkes models with varying productivities to each of the 50 United States individually, estimating for each state a transmission time density, both nonparametrically and using a normal approximation. We find that, for nearly all states, the estimated transmission times are centered near seven days with a standard deviation of approximately one day. Compared to previous reports, the results here suggest that transmission times for SARS-CoV-2 are somewhat shorter, on average, and the distribution is less diffuse, though the results also suggest the possibility of transmission occurring on the first day of exposure.

The characterisation of future extreme wave events is crucial because of their multiple impacts, covering a broad range of topics such as coastal flood hazard, coastal erosion, reliability of offshore and coastal structures. The main goal of this paper is to propose and study a stochastic simulator that, given offshore conditions (peak direction ${\mathit{D}}_{\mathit{p}}$, peak period ${\mathit{T}}_{\mathit{p}}$ and moderately high significant wave heights ${\mathit{H}}_{\mathit{s}}$), produces jointly offshore and coastal extreme ${\mathit{H}}_{\mathit{s}}$, a quantity measuring the wave severity and which represent a key feature in coastal risk analysis. For this purpose we rely on bivariate Peaks over Threshold, and a nonparametric simulation scheme of bivariate GPD is developed. From this joint simulator, a second generator is derived, allowing for conditional simulations of extreme ${\mathit{H}}_{\mathit{s}}$. Finally, to take into account nonstationarities, the extended generalised Pareto model is also adapted, letting the parameters vary with specific sea-state parameters ${\mathit{T}}_{\mathit{p}}$ and ${\mathit{D}}_{\mathit{p}}$. The performances of the two proposed generators are illustrated on simulated data and then applied to the simulation of new extreme oceanographic conditions close to the French Brittany coast using hindcast sea-state data. Results show that the proposed algorithms successfully simulate future extreme ${\mathit{H}}_{\mathit{s}}$ near the coast in a nonparametric way, jointly or conditionally on sea-state parameters from a coarser model.

Individualized treatment rules (ITRs) for treatment recommendation is an important topic for precision medicine as not all beneficial treatments work well for all individuals. Interpretability is a desirable property of ITRs, as it helps practitioners make sense of treatment decisions, yet there is a need for ITRs to be flexible to effectively model complex biomedical data for treatment decision making. Many ITR approaches either focus on linear ITRs, which may perform poorly when true optimal ITRs are nonlinear, or black-box nonlinear ITRs, which may be hard to interpret and can be overly complex. This dilemma indicates a tension between interpretability and accuracy of treatment decisions. Here we propose an additive model-based nonlinear ITR learning method that balances interpretability and flexibility of the ITR. Our approach aims to strike this balance by allowing both linear and nonlinear terms of the covariates in the final ITR. Our approach is parsimonious in that the nonlinear term is included in the final ITR only when it substantially improves the ITR performance. To prevent overfitting, we combine crossfitting and a specialized information criterion for model selection. Through extensive simulations we show that our methods are data-adaptive to the degree of nonlinearity and can favorably balance ITR interpretability and flexibility. We further demonstrate the robust performance of our methods with an application to a cancer drug sensitive study.

Multimodel ensemble analysis integrates information from multiple climate models into a unified projection. However, existing integration approaches, based on model averaging, can dilute fine-scale spatial information and incur bias from rescaling low-resolution climate models. We propose a statistical approach, called NN-GPR, using Gaussian process regression (GPR) with an infinitely wide deep neural network based covariance function. NN-GPR requires no assumptions about the relationships between climate models, no interpolation to a common grid, and automatically downscales as part of its prediction algorithm. Model experiments show that NN-GPR can be highly skillful at surface temperature and precipitation forecasting by preserving geospatial signals at multiple scales and capturing interannual variability. Our projections particularly show improved accuracy and uncertainty quantification skill in regions of high variability, which allows us to cheaply assess tail behavior at a 0.44${}^{\circ }$/50 km spatial resolution without a regional climate model (RCM). Evaluations on reanalysis data and SSP2-4.5 forced climate models show that NN-GPR produces similar, overall climatologies to the model ensemble while better capturing fine-scale spatial patterns. Finally, we compare NN-GPR’s regional predictions against two RCMs and show that NN-GPR can rival the performance of RCMs using only global model data as input.

Categorizing individual cells into one of many known cell-type categories, also known as cell-type annotation, is a critical step in the analysis of single-cell genomics data. The current process of annotation is time intensive and subjective, which has led to different studies describing cell types with labels of varying degrees of resolution. While supervised learning approaches have provided automated solutions to annotation, there remains a significant challenge in fitting a unified model for multiple datasets with inconsistent labels. In this article we propose a new multinomial logistic regression estimator which can be used to model cell-type probabilities by integrating multiple datasets with labels of varying resolution. To compute our estimator, we solve a nonconvex optimization problem using a blockwise proximal gradient descent algorithm. We show through simulation studies that our approach estimates cell-type probabilities more accurately than competitors in a wide variety of scenarios. We apply our method to 10 single-cell RNA-seq datasets and demonstrate its utility in predicting fine resolution cell-type labels on unlabeled data as well as refining cell-type labels on data with existing coarse resolution annotations. Finally, we demonstrate that our method can lead to novel scientific insights in the context of a differential expression analysis comparing peripheral blood gene expression before and after treatment with interferon-β. An R package implementing the method is available in the Supplementary Material and at https://github.com/keshav-motwani/IBMR, and the collection of datasets we analyze is available at https://github.com/keshav-motwani/AnnotatedPBMC.

Differential coexpression analysis has been widely applied by scientists in understanding the biological mechanisms of diseases. However, the unknown differential patterns are often complicated; thus, models based on simplified parametric assumptions can be ineffective in identifying the differences. Meanwhile, the gene expression data involved in such analysis are in extremely high dimensions by nature, whose correlation matrices may not even be computable. Such a large scale seriously limits the application of most well-studied statistical methods. This paper introduces a simple yet powerful approach to the differential correlation analysis problem called compressed spectral screening. By leveraging spectral structures and random sampling techniques, our approach could achieve a highly accurate screening of features with complicated differential patterns while maintaining the scalability to analyze correlation matrices of ${10}^4$–${10}^5$ variables within a few minutes on a standard personal computer. We have applied this screening approach in comparing a TCGA data set about Glioblastoma with normal subjects. Our analysis successfully identifies multiple functional modules of genes that exhibit different coexpression patterns. The findings reveal new insights about Glioblastoma’s evolving mechanism. The validity of our approach is also justified by a theoretical analysis, showing that the compressed spectral analysis can achieve variable screening consistency.

Determining the adsorption isotherms is an issue of significant importance in preparative chromatography. A modern technique for estimating adsorption isotherms is to solve an inverse problem so that the simulated batch separation coincides with actual experimental results. However, due to the ill-posedness, the high nonlinearity, and the uncertainty quantification of the corresponding physical model, the existing deterministic inversion methods are usually inefficient in real-world applications. To overcome these difficulties and study the uncertainties of the adsorption-isotherm parameters, in this work, based on the Bayesian sampling framework, we propose a statistical approach for estimating the adsorption isotherms in various chromatography systems. Two modified Markov chain Monte Carlo algorithms are developed for a numerical realization of our statistical approach. Numerical experiments with both synthetic and real data are conducted and described to show the efficiency of the proposed new method.

Reliable estimates of sea-level return-levels are crucial for coastal flooding risk assessments and for coastal flood defence design. We describe a novel method for estimating extreme sea-levels that is the first to capture seasonality, interannual variations and longer term changes. We use a joint probabilities method, with skew-surge and peak-tide as two sea-level components. The tidal regime is predictable, but skew-surges are stochastic. We present a statistical model for skew-surges, where the main body of the distribution is modelled empirically while a nonstationary generalised Pareto distribution (GPD) is used for the upper tail. We capture within-year seasonality by introducing a daily covariate to the GPD model and allowing the distribution of peak-tide to change over months and years. Skew-surge-peak-tide dependence is accounted for, via a tidal covariate, in the GPD model, and we adjust for skew-surge temporal dependence through the subasymptotic extremal index. We incorporate spatial prior information in our GPD model to reduce the uncertainty associated with the highest return-level estimates. Our results are an improvement on current return-level estimates, with previous methods typically underestimating. We illustrate our method at four U.K. tide gauges.

A large literature on judicial decision making asks if judges with different features of an attribute (e.g, sex, race) adjudicate cases differently. Researchers estimate models for case outcomes, interpreting coefficients associated with attributes as effects. But attributes are not treatments. While these coefficients indicate how judges with different features adjudicate the different cases they are assigned, ideally, different judges should be compared on a common set of cases. We construct a general methodology for making such comparisons, using it to study whether monetary relief in discrimination cases brought by the Equal Employment Opportunity Commission differs by judges’ race. For all federal judges (treatments) eligible to hear a case (unit), we define potential outcomes, using unit treatment effects between judges with different features to define a unit feature comparison (UFC), then using these to define new population estimands: the average (AFC) and quantile (QFC) feature comparisons. We estimate these quantities by combining observed case outcomes with missing potential outcomes imputed from the posterior predictive distribution of a two-part Bayesian hierarchical model. A case initially assigned to a non-white or African American judge is more likely to result in monetary relief than were that case initially assigned to an eligible white or non-African American judge. For the amount of relief, the 95% posterior interval for the AFC covers 0, while the upper endpoint of the 95% posterior interval for the median QFC is negative.

The Scientific Registry of Transplant Recipients (SRTR) system has become a rich resource for understanding the complex mechanisms of graft failure after kidney transplant, a crucial step for allocating organs effectively and implementing appropriate care. As transplant centers that treated patients might strongly confound graft failures, Cox models stratified by centers can eliminate their confounding effects. Also, since recipient age is a proven nonmodifiable risk factor, a common practice is to fit models separately by recipient age groups. The moderate sample sizes, relative to the number of covariates, in some age groups may lead to biased maximum stratified partial likelihood estimates and unreliable confidence intervals, even when samples still outnumber covariates. To draw reliable inference on a comprehensive list of risk factors measured from both donors and recipients in SRTR, we propose a debiased lasso approach via quadratic programming for fitting stratified Cox models. We establish asymptotic properties and verify via simulations that our method produces consistent estimates and confidence intervals with nominal coverage probabilities. Accounting for nearly 100 confounders in SRTR, the debiased method detects that the graft failure hazard nonlinearly increases with donor’s age among all recipient age groups and that organs from older donors more adversely impact the younger recipients. Our method also delineates the associations between graft failure and many risk factors such as recipients’ primary diagnoses (e.g., polycystic disease, glomerular disease, and diabetes) and donor-recipient mismatches for human leukocyte antigen loci across recipient age groups. These results may inform the refinement of donor-recipient matching criteria for stakeholders.

There is great interest in ecology to understand how wild animals are affected by anthropogenic disturbances, such as sounds. For example, behavioural response studies are an important approach to quantify the impact of naval activity on marine mammals. Controlled exposure experiments are undertaken where the behaviour of animals is quantified before, during, and after exposure to a controlled sound source, often using telemetry tags (e.g., accelerometers or satellite trackers). Statistical modelling is required to formally compare patterns before and after exposure, to quantify deviations from baseline behaviour. We propose varying-coefficient stochastic differential equations (SDEs) as a flexible framework to model such data with two components: (1) time-varying baseline dynamics, modelled with nonparametric or random effects of time-varying covariates, and (2) a nonparametric response model, which captures deviations from baseline. SDEs are specified in continuous time, which makes it straightforward to analyse data collected at irregular time intervals, a common situation for animal tracking studies. We describe how the model can be embedded into a state-space modelling framework to account for measurement error. We present inferential methods for model fitting, model checking, and uncertainty quantification (including on the response model). We apply this approach to two behavioural response study data sets on beaked whales: a satellite track and high-resolution depth data. Our results suggest that the whales’ horizontal movement and vertical diving behaviour changed after exposure to the sound source, and future work should evaluate the severity and possible consequences of these responses. These two very different examples showcase the versatility of varying-coefficient SDEs to measure changes in behaviour, and we discuss implications of disturbances for the whales’ energetic balance.