Data Setting Research Articles

Optimizing Multi-Objective Federated Learning on Non-IID Data with Improved NSGA-III and Hierarchical Clustering

Federated learning (FL) can tackle the problem of data silos of asymmetric information and privacy leakage; however, it still has shortcomings, such as data heterogeneity, high communication cost and uneven distribution of performance. To overcome these issues and achieve parameter optimization of FL on non-Independent Identically Distributed (non-IID) data, a multi-objective FL parameter optimization method based on hierarchical clustering and the third-generation non-dominated sorted genetic algorithm III (NSGA-III) algorithm is proposed, which aims to simultaneously minimize the global model error rate, global model accuracy distribution variance and communication cost. The introduction of a hierarchical clustering algorithm on non-IID data can accelerate convergence so that FL can employ an evolutionary algorithm with a low FL client participation ratio, reducing the overall communication cost of the NSGA-III algorithm. Meanwhile, the NSGA-III algorithm, with fast greedy initialization and a strategy of discarding low-quality individuals (named NSGA-III-FD), is proposed to improve the convergence efficiency and the quality of Pareto-optimal solutions. Under two non-IID data settings, the CNN experiments on both MNIST and CIFAR-10 datasets show that our approach can obtain better Pareto-optimal solutions than classical evolutionary algorithms, and the selected solutions with an optimized model can achieve better multi-objective equilibrium than the standard federated averaging (FedAvg) algorithm and the Clustering-based FedAvg algorithm.

Inclusion of flood diversion canal operation in the H08 hydrological model with a case study from the Chao Phraya River basin: model development and validation

Abstract. Water diversion systems play crucial roles in assuaging flood risk by diverting and redistributing water within and among basins. For flood and drought assessments, including investigations of the effects of diversion systems on river discharge worldwide, the explicit inclusion of these systems into global hydrological models (GHMs) is essential. However, such representation remains in the pioneering stage because of complex canal operations and insufficient data. Therefore, we developed a regionalized canal operation scheme and implemented it in the H08 GHM for flood diversion in the Chao Phraya River basin (CPRB), Thailand, which is a complex river network with several natural and artificial diversion canals and has been subject to severe flooding in the past, including recent years. Region-specific validation results revealed that the enhanced H08 model with the regionalized diversion scheme could effectively simulate the observed flood diversion pattern in the CPRB. Diverted water comprises approximately 49 % of the annual average river discharge in the CPRB. The simulations further confirmed that the presented canal scheme had the potential to reduce flood risk in the basin by significantly reducing the number of flooding days. A generalized canal scheme with simple input data settings was also constructed for future global applications, providing insights into the maximum level of discharge reduction achievable with diversion of nearly 57 % of the annual average river discharge of the CPRB. Overall, the enhanced H08 model with canal schemes can be adapted and applied to different contexts and regions, accounting for the characteristics of each river network by maintaining the basic principles unaltered.

Local polynomial smoothing based on the Kaplan–Meier estimate

The local polynomial modeling of the Kaplan–Meier estimate for random designs under the right censored data setting is investigated in detail. Two classes of boundary aware estimates are developed: estimates of the distribution function and its derivatives of any arbitrary order and estimates of integrated distribution function derivative products. Their statistical properties are quantified analytically and their implementation is facilitated by the development of corresponding data driven plug-in bandwidth selectors. The asymptotic rate of convergence of the plug-in rule for the estimates of the distribution function and its derivatives is quantified analytically. Numerical evidence is also provided on its finite sample performance. A real life data analysis illustrates how the methodological advances proposed herein help to generate additional insights in comparison to existing methods.

Twenty years since Joinpoint 1.0: Two major enhancements, their justification, and impact.

Since its release of Version 1.0 in 1998, Joinpoint software developed for cancer trend analysis by a team at the US National Cancer Institute has received a considerable attention in the trend analysis community and it became one of most widely used software for trend analysis. The paper published in Statistics in Medicine in 2000 (a previous study) describes the permutation test procedure to select the number of joinpoints, and Joinpoint Version 1.0 implemented the permutation procedure as the default model selection method and employed parametric methods for the asymptotic inference of the model parameters. Since then, various updates and extensions have been made in Joinpoint software. In this paper, we review basic features of Joinpoint, summarize important updates of Joinpoint software since its first release in 1998, and provide more information on two major enhancements. More specifically, these enhancements overcome prior limitations in both the accuracy and computational efficiency of previously used methods. The enhancements include: (i) data driven model selection methods which are generally more accurate under a broad range of data settings and more computationally efficient than the permutation test and (ii) the use of the empirical quantile method for construction of confidence intervals for the slope parameters and the location of the joinpoints, which generally provides more accurate coverage than the prior parametric methods used. We show the impact of these changes in cancer trend analysis published by the US National Cancer Institute.

Modeling to inform economy‐wide pandemic policy: Bringing epidemiologists and economists together

Modeling to inform economy‐wide pandemic policy: Bringing epidemiologists and economists together

Sample size methods for evaluation of predictive biomarkers.

Treatment selection biomarkers are those that can be useful in guiding choice of therapy. Just as new therapies require evaluation in appropriately designed clinical trials to determine their benefit, therapy selection biomarkers require evaluation in appropriately designed studies. These studies may be prospective clinical trials or retrospective studies based on specimens stored from a completed clinical trial. Ideally, patient treatment assignments should be randomized, and consideration should be given to an appropriate sample size-either for prospective planning of a new study or access to a sufficient number of stored specimens. Here, we develop a novel sample size method for estimation of a confidence interval of specified average width, for an intuitively appealing previously proposed parameter that reflects the expected benefit of using biomarker-guided therapy relative to a standard-of-care therapy. The estimation approach combines Monte Carlo and regression to result in a procedure that performs well over a range of scenarios. Although derived under a specific Cox proportional hazards regression model, robustness to model violations is demonstrated by evaluation under accelerated failure time and cure models. The sample size method produces adequate or conservative sample size estimates under a range of scenarios. Computer code in R and C++, and applications for Mac and Windows are made available for implementation of the sample size estimation procedure. The method is applied to a real data setting and results discussed.

Anderson relaxation test for intrinsic dimension selection in model-based clustering

Parsimonious finite mixture models often require the a priori selection of desired model dimensionality. For example, projection-based parsimonious models demand the dimension of the subspace for projection. Other models ask for their own structural restrictions on parameters. The subspace clustering framework is a projection-based parsimonious model for various finite mixtures, including the Gaussian variant. The existing dimension selection methods for subspace clustering are ad-hoc or potentially computationally prohibitive, creating a need for a principled, yet computationally lightweight, approach. In light of this problem, a hypothesis test-based intrinsic dimension estimation method called the Anderson Relaxation Test (ART) is introduced, and its performance is examined in both simulated and real data settings.

Prehospital Buprenorphine Treatment for Opioid Use Disorder by Paramedics: First Year Results of the EMS Buprenorphine Use Pilot

Background Prehospital initiation of buprenorphine treatment for Opioid Use Disorder (OUD) by paramedics is an emerging potential intervention to reach patients at greatest risk for opioid-related death. Emergency medical services (EMS) patients who are at high risk for overdose deaths may never engage in treatment as they frequently refuse transport to the hospital after naloxone reversal. The potentially important role of EMS as the initiator for medication for opioid use disorder (MOUD) in the most high-risk patients has not been well described. Setting This project relies on four interventions: a public access naloxone distribution program, an electronic trigger and data sharing program, an “Overdose Receiving Center,” and a paramedic-initiated buprenorphine treatment. For the final intervention, paramedics followed a protocol-based pilot that had an EMS physician consultation prior to administration. Results There were 36 patients enrolled in the trial study in the first year who received buprenorphine. Of those patients receiving buprenorphine, only one patient signed out against medical advice on scene. All other patients were transported to an emergency department and their clinical outcome and 7 and 30 day follow ups were determined by the substance use navigator (SUN). Thirty-six of 36 patients had follow up data obtained in the short term and none experienced any precipitated withdrawal or other adverse outcomes. Patients had a 50% (18/36) rate of treatment retention at 7 days and 36% (14/36) were in treatment at 30 days. Conclusion In this small pilot project, paramedic-initiated buprenorphine in the setting of data sharing and linkage with treatment appears to be a safe intervention with a high rate of ongoing outpatient treatment for risk of fatal opioid overdoses.

Transferring model structure in Bayesian transfer learning for Gaussian process regression

Bayesian transfer learning (BTL) is defined in this paper as the task of conditioning a target probability distribution on a transferred source distribution. The target globally models the interaction between the source and target, and conditions on a probabilistic data predictor made available by an independent local source modeller. Fully probabilistic design is adopted to solve this optimal decision-making problem in the target. By successfully transferring higher moments of the source, the target can reject unreliable source knowledge (i.e. it achieves robust transfer). This dual-modeller framework means that the source’s local processing of raw data into a transferred predictive distribution – with compressive possibilities – is enriched by (the possible expertise of) the local source model. In addition, the introduction of the global target modeller allows correlation between the source and target tasks – if known to the target – to be accounted for. Important consequences emerge. Firstly, the new scheme attains the performance of fully modelled (i.e. conventional) multitask learning schemes in (those rare) cases where target model misspecification is avoided. Secondly, and more importantly, the new dual-modeller framework is robust to the model misspecification that undermines conventional multitask learning. We thoroughly explore these issues in the key context of interacting Gaussian process regression tasks. Experimental evidence from both synthetic and real data settings validates our technical findings: that the proposed BTL framework enjoys robustness in transfer while also being robust to model misspecification.

High-dose Radiotherapy or Androgen Deprivation Therapy (HEAT) as Treatment Intensification for Localized Prostate Cancer: An Individual Patient–data Network Meta-analysis from the MARCAP Consortium

BackgroundThe relative benefits of radiotherapy (RT) dose escalation and the addition of short-term or long-term androgen deprivation therapy (STADT or LTADT) in the treatment of prostate cancer are unknown. ObjectiveTo perform a network meta-analysis (NMA) of relevant randomized trials to compare the relative benefits of RT dose escalation ± STADT or LTADT. Design, setting, and participantsAn NMA of individual patient data from 13 multicenter randomized trials was carried out for a total of 11862 patients. Patients received one of the six permutations of low-dose RT (64 to <74 Gy) ± STADT or LTADT, high-dose RT (≥74 Gy), or high-dose RT ± STADT or LTADT. Outcome measurements and statistical analysesMetastasis-free survival (MFS) was the primary endpoint. Frequentist and Bayesian NMAs were performed to rank the various treatment strategies by MFS and biochemical recurrence–free survival (BCRFS). Results and limitationsMedian follow-up was 8.8 yr (interquartile range 5.7–11.5). The greatest relative improvement in outcomes was seen for addition of LTADT, irrespective of RT dose, followed by addition of STADT, irrespective of RT dose. RT dose escalation did not improve MFS either in the absence of ADT (hazard ratio [HR] 0.97, 95% confidence interval [CI] 0.80–1.18) or with STADT (HR 0.99, 95% CI 0.8–1.23) or LTADT (HR 0.94, 95% CI 0.65–1.37). According to P-score ranking and rankogram analysis, high-dose RT + LTADT was the optimal treatment strategy for both BCRFS and longer-term outcomes. ConclusionsConventionally escalated RT up to 79.2 Gy, alone or in the presence of ADT, does not improve MFS, while addition of STADT or LTADT to RT alone, regardless of RT dose, consistently improves MFS. RT dose escalation does provide a high probability of improving BCRFS and, provided it can be delivered without compromising quality of life, may represent the optimal treatment strategy when used in conjunction with ADT. Patient summaryUsing a higher radiotherapy dose when treating prostate cancer does not reduce the chance of developing metastases or death, but it does reduce the chance of having a rise in prostate-specific antigen (PSA) signifying recurrence of cancer. Androgen deprivation therapy improves all outcomes. A safe increase in radiotherapy dose in conjunction with androgen deprivation therapy may be the optimal treatment.

Robust and accurate estimation of cellular fraction from tissue omics data via ensemble deconvolution.

Tissue-level omics data such as transcriptomics and epigenomics are an average across diverse cell types. To extract cell-type-specific (CTS) signals, dozens of cellular deconvolution methods have been proposed to infer cell-type fractions from tissue-level data. However, these methods produce vastly different results under various real data settings. Simulation-based benchmarking studies showed no universally best deconvolution approaches. There have been attempts of ensemble methods, but they only aggregate multiple single-cell references or reference-free deconvolution methods. To achieve a robust estimation of cellular fractions, we proposed EnsDeconv (Ensemble Deconvolution), which adopts CTS robust regression to synthesize the results from 11 single deconvolution methods, 10 reference datasets, 5 marker gene selection procedures, 5 data normalizations and 2 transformations. Unlike most benchmarking studies based on simulations, we compiled four large real datasets of 4937 tissue samples in total with measured cellular fractions and bulk gene expression from different tissues. Comprehensive evaluations demonstrated that EnsDeconv yields more stable, robust and accurate fractions than existing methods. We illustrated that EnsDeconv estimated cellular fractions enable various CTS downstream analyses such as differential fractions associated with clinical variables. We further extended EnsDeconv to analyze bulk DNA methylation data. EnsDeconv is freely available as an R-package from https://github.com/randel/EnsDeconv. The RNA microarray data from the TRAUMA study are available and can be accessed in GEO (GSE36809). The demographic and clinical phenotypes can be shared on reasonable request to the corresponding authors. The RNA-seq data from the EVAPR study cannot be shared publicly due to the privacy of individuals that participated in the clinical research in compliance with the IRB approval at the University of Pittsburgh. The RNA microarray data from the FHS study are available from dbGaP (phs000007.v32.p13). The RNA-seq data from ROS study is downloaded from AD Knowledge Portal. Supplementary data are available at Bioinformatics online.

Towards a robust out-of-the-box neural network model for genomic data

BackgroundThe accurate prediction of biological features from genomic data is paramount for precision medicine and sustainable agriculture. For decades, neural network models have been widely popular in fields like computer vision, astrophysics and targeted marketing given their prediction accuracy and their robust performance under big data settings. Yet neural network models have not made a successful transition into the medical and biological world due to the ubiquitous characteristics of biological data such as modest sample sizes, sparsity, and extreme heterogeneity.ResultsHere, we investigate the robustness, generalization potential and prediction accuracy of widely used convolutional neural network and natural language processing models with a variety of heterogeneous genomic datasets. Mainly, recurrent neural network models outperform convolutional neural network models in terms of prediction accuracy, overfitting and transferability across the datasets under study.ConclusionsWhile the perspective of a robust out-of-the-box neural network model is out of reach, we identify certain model characteristics that translate well across datasets and could serve as a baseline model for translational researchers.

An algorithm for distributed Bayesian inference

Monte Carlo algorithms, such as Markov chain Monte Carlo (MCMC) and Hamiltonian Monte Carlo (HMC), are routinely used for Bayesian inference; however, these algorithms are prohibitively slow in massive data settings because they require multiple passes through the full data in every iteration. Addressing this problem, we develop a scalable extension of these algorithms using the divide‐and‐conquer (D&amp;C) technique that divides the data into a sufficiently large number of subsets, draws parameters in parallel on the subsets using a powered likelihood and produces Monte Carlo draws of the parameter by combining parameter draws obtained from each subset. The combined parameter draws play the role of draws from the original sampling algorithm. Our main contributions are twofold. First, we demonstrate through diverse simulated and real data analyses focusing on generalized linear models (GLMs) that our distributed algorithm delivers comparable results as the current state‐of‐the‐art D&amp;C algorithms in terms of statistical accuracy and computational efficiency. Second, providing theoretical support for our empirical observations, we identify regularity assumptions under which the proposed algorithm leads to asymptotically optimal inference. We also provide illustrative examples focusing on normal linear and logistic regressions where parts of our D&amp;C algorithm are analytically tractable.

Multivariate functional group sparse regression: Functional predictor selection.

In this paper, we propose methods for functional predictor selection and the estimation of smooth functional coefficients simultaneously in a scalar-on-function regression problem under a high-dimensional multivariate functional data setting. In particular, we develop two methods for functional group-sparse regression under a generic Hilbert space of infinite dimension. We show the convergence of algorithms and the consistency of the estimation and the selection (oracle property) under infinite-dimensional Hilbert spaces. Simulation studies show the effectiveness of the methods in both the selection and the estimation of functional coefficients. The applications to functional magnetic resonance imaging (fMRI) reveal the regions of the human brain related to ADHD and IQ.

Deep dynamic modeling with just two time points: Can we still allow for individual trajectories?

Longitudinal biomedical data are often characterized by a sparse time grid and individual-specific development patterns. Specifically, in epidemiological cohort studies and clinical registries we are facing the question of what can be learned from the data in an early phase of the study, when only a baseline characterization and one follow-up measurement are available. Inspired by recent advances that allow to combine deep learning with dynamic modeling, we investigate whether such approaches can be useful for uncovering complex structure, in particular for an extreme small data setting with only two observations time points for each individual. Irregular spacing in time could then be used to gain more information on individual dynamics by leveraging similarity of individuals. We provide a brief overview of how variational autoencoders (VAEs), as a deep learning approach, can be linked to ordinary differential equations(ODEs) for dynamic modeling, and then specifically investigate the feasibility of such an approach that infers individual-specific latent trajectories by including regularity assumptions and individuals' similarity. We also provide a description of this deep learning approach as a filtering task to give a statistical perspective. Using simulated data, we show to what extent the approach can recover individual trajectories from ODE systems with two and four unknown parameters and infer groups of individuals with similar trajectories, and where it breaks down. The results show that such dynamic deep learning approaches can be useful even in extreme small data settings, but need to be carefullyadapted.

A comprehensive evaluation of regression-based drug responsiveness prediction models, using cell viability inhibitory concentrations (IC50 values).

Predicting drug response is critical for precision medicine. Diverse methods have predicted drug responsiveness, as measured by the half-maximal drug inhibitory concentration (IC50), in cultured cells. Although IC50s are continuous, traditional prediction models have dealt mainly with binary classification of responsiveness. However, since there are few regression-based IC50 predictions, comprehensive evaluations of regression-based IC50 prediction models, including machine learning (ML) and deep learning (DL), for diverse data types and dataset sizes, have not been addressed. Here, we constructed 11 input data settings, including multi-omics settings, with varying dataset sizes, then evaluated the performance of regression-based ML and DL models to predict IC50s. DL models considered two convolutional neural network architectures: CDRScan and residual neural network (ResNet). ResNet was introduced in regression-based DL models for predicting drug response for the first time. As a result, DL models performed better than ML models in all the settings. Also, ResNet performed better than or comparable to CDRScan and ML models in all settings. The data underlying this article are available in GitHub at https://github.com/labnams/IC50evaluation. Supplementary data are available at Bioinformatics online.

Explicit Calculations of Wet‐Bulb Globe Temperature Compared With Approximations and Why It Matters for Labor Productivity

AbstractWet‐bulb globe temperature (WBGT) is a widely applied heat stress index. However, most applications of WBGT within the heat stress impact literature that do not use WBGT at all, but use one of the ad hoc approximations, typically the simplified WBGT (sWBGT) or the environmental stress index (ESI). Surprisingly, little is known about how well these approximations work for the global climate and climate change settings that they are being applied to. Here, we assess the bias distribution as a function of temperature, humidity, wind speed, and radiative conditions of both sWBGT and ESI relative to a well‐validated, explicit physical model for WBGT developed by Liljegren, within an idealized context and the more realistic setting of ERA5 reanalysis data. sWBGT greatly overestimates heat stress in hot‐humid areas. ESI has much smaller biases in the range of standard climatological conditions. Over subtropical dry regions, both metrics can substantially underestimate extreme heat. We show systematic overestimation of labor loss by sWBGT over much of the world today. We recommend discontinuing the use of sWBGT. ESI may be acceptable for assessing average heat stress or integrated impact over a long period like a year, but less suitable for health applications, extreme heat stress analysis, or as an operational index for heat warning, heatwave forecasting, or guiding activity modification at the workplace. Nevertheless, Liljegren's approach should be preferred over these ad hoc approximations and we provide a fast Python implementation to encourage its widespread use.

Data-Driven Flood Alert System (FAS) Using Extreme Gradient Boosting (XGBoost) to Forecast Flood Stages

Heavy rainfall leads to severe flooding problems with catastrophic socio-economic impacts worldwide. Hydrologic forecasting models have been applied to provide alerts of extreme flood events and reduce damage, yet they are still subject to many uncertainties due to the complexity of hydrologic processes and errors in forecasted timing and intensity of the floods. This study demonstrates the efficacy of using eXtreme Gradient Boosting (XGBoost) as a state-of-the-art machine learning (ML) model to forecast gauge stage levels at a 5-min interval with various look-out time windows. A flood alert system (FAS) built upon the XGBoost models is evaluated by two historical flooding events for a flood-prone watershed in Houston, Texas. The predicted stage values from the FAS are compared with observed values with demonstrating good performance by statistical metrics (RMSE and KGE). This study further compares the performance from two scenarios with different input data settings of the FAS: (1) using the data from the gauges within the study area only and (2) including the data from additional gauges outside of the study area. The results suggest that models that use the gauge information within the study area only (Scenario 1) are sufficient and advantageous in terms of their accuracy in predicting the arrival times of the floods. One of the benefits of the FAS outlined in this study is that the XGBoost-based FAS can run in a continuous mode to automatically detect floods without requiring an external starting trigger to switch on as usually required by the conventional event-based FAS systems. This paper illustrates a data-driven FAS framework as a prototype that stakeholders can utilize solely based on their gauging information for local flood warning and mitigation practices.

Protected attribute guided representation learning for bias mitigation in limited data

In many real-world problem settings, the training data may contain certain unwanted correlations or biases. Such biases generally occur due to imbalances/skew in the data with respect to some protected attributes. When a standard deep learning model is trained on such biased data, the model will also learn these biases and produce biased predictions. If the available training data is limited, then the impact of biases on the model predictions is magnified, and the performance of existing bias mitigation strategies is also severely affected. We propose a novel approach for mitigating bias in an image classification model at the representation level in the limited data setting. Specifically, we use the protected attribute to guide our representation learning process in the limited data setting such that the learned feature space becomes robust to the biases occurring due to the protected attribute, thereby reducing the impact of bias in the model predictions. We experimentally demonstrate that our proposed bias mitigation approach significantly reduces the model bias score and improves the model performance on seven benchmark datasets. We also demonstrate that our proposed method is also complementary to existing bias mitigation approaches and significantly reduces the bias score of the baseline, sampling, adversarial, domain discriminative, and domain-independent models by 89.97%, 90.94%, 86.74%, 83.62%, and 62.96%, respectively, on the reduced CIFAR-10S dataset. It also significantly improves the mean accuracy of the baseline, sampling, adversarial, domain discriminative, and domain-independent models by absolute margins of 18.83%, 8.32%, 11.78%, 18.74%, and 7.13%, respectively, on reduced CIFAR-10S.

An Augmented High-Dimensional Graphical Lasso Method to Incorporate Prior Biological Knowledge for Global Network Learning.

Biological networks are often inferred through Gaussian graphical models (GGMs) using gene or protein expression data only. GGMs identify conditional dependence by estimating a precision matrix between genes or proteins. However, conventional GGM approaches often ignore prior knowledge about protein-protein interactions (PPI). Recently, several groups have extended GGM to weighted graphical Lasso (wGlasso) and network-based gene set analysis (Netgsa) and have demonstrated the advantages of incorporating PPI information. However, these methods are either computationally intractable for large-scale data, or disregard weights in the PPI networks. To address these shortcomings, we extended the Netgsa approach and developed an augmented high-dimensional graphical Lasso (AhGlasso) method to incorporate edge weights in known PPI with omics data for global network learning. This new method outperforms weighted graphical Lasso-based algorithms with respect to computational time in simulated large-scale data settings while achieving better or comparable prediction accuracy of node connections. The total runtime of AhGlasso is approximately five times faster than weighted Glasso methods when the graph size ranges from 1,000 to 3,000 with a fixed sample size (n = 300). The runtime difference between AhGlasso and weighted Glasso increases when the graph size increases. Using proteomic data from a study on chronic obstructive pulmonary disease, we demonstrate that AhGlasso improves protein network inference compared to the Netgsa approach by incorporating PPI information.