Estimation Of Model Coefficients Research Articles

1322. Quantifying the Effects of Frequently Prescribed Antimicrobials with Perceived Potential for QT Interval Prolongation during the COVID-19 Era

BackgroundCountless diseases and medications have been implicated in the past as causing prolongation of the QT interval. Their unique role through the means of quantifying the definite magnitude of relative risk they contribute during hospitalization still requires further investigation. The aim of this study was to describe the impact of commonly used anti-infectives on the QT interval in hospitalized patients during the COVID-19 era.MethodsDemographic information, medical history, laboratory data, medication administration history and ECG recording data was collected from the electronic records of adult patients admitted to two urban hospitals. A mixed effects approach with four sub-models for the QT interval comprised of: heart rate, circadian rhythm, gender, and the drug (regressed as the cumulative mg dose administered over time) and disease effects was used. Fixed and random effects with between occasion variability were estimated for the parameters with a Bayesian approach using the STAN software.ResultsData from 2180 patients were used with baseline characteristics shown in Table 1. Observed vs. predicted plots based on the training (Figure 1.A) and validation data set (Figure 1.B) showed excellent fit. The parameters for QTc0, α, gender, and circadian rhythm were identified within the range previously described (Table 2.). Similarly, the model correctly identified the impact of acute or chronic diseases on the QT interval. Model coefficient estimates [mean (95% CI) of 0.010 (0.006, 0.15) and 0.0045 (0.0013, 0.01100) msec/mg cumulative dose, respectively] suggest that patients treated with conventional regimens of fluconazole and levofloxacin are most likely to present with a QT interval increase > 5 msec, the cutoff threshold of regulatory concern.Figure 1. A-B Table 1. Table 2. ConclusionThe model developed accurately identified the impact baseline risk factors and concomitant medications have on the QT interval. When adjusted for these confounding variables, estimates of QT interval prolongation show that treatment with fluconazole and levofloxacin pose a considerable risk; while treatment with azithromycin or hydroxychloroquine is of moderate risk for QT interval prolongation.Disclosures All Authors: No reported disclosures

A simulation analysis to study the temporal and spatial aggregations of safety datasets with excess zero observations

Crash data are often characterized with numerous zero observations. Sometimes, the number of zero observations is directly correlated with the selected spatial and/or temporal scales for data aggregation. Finding a balance in aggregation is a critical task in data preparation. On the one hand, using the disaggregated data may result in having excessive zero observations, in which the popular negative binomial model may not be adequate for the safety analysis. On the other hand, too much aggregation may result in loss of information. This paper documents a simulation study that aimed at determining criteria for deciding when data aggregation is needed. The simulation study explores the information loss due to aggregation as a function of precision or accuracy in estimation of model coefficients. The simulation results indicate that the reduction in variability, i.e. coefficient of variation, of the independent variables after aggregation is important criteria to decide on the aggregation level.

Protection of six-phase transmission line using recursive estimation of non-linear autoregression model coefficients and decision tree

Ever-increasing energy demands and limited right of way over land motivates the adoption of six-phase power transmission. However, the wider recognition of six-phase transmission necessitates the development of a reliable protection system. Soaring assimilation of non-linear loads has complicated the protection task due to the intrusion of harmonics. This work aims at imparting necessary protection to six-phase transmission systems with immunity against harmonics. The issue pertaining to harmonics and sensor noise-induced non-linearity in the signal being fed to the protective relays has been addressed by modelling the current–voltage profile using a non-linear autoregression approach. Conventional relays are known to malfunction during non-linear loading and noisy measurement. In this context, an adaptive window-based approach has been employed to capture the non-linear dynamics, followed by mapping the model coefficients with the state of the six-phase system using decision trees. The inclusion of parameters in the non-linear auto regressive model for representing harmonic intrusion and noise imparts necessary robustness to the proposed protection scheme against non-linear loading and noisy sensor measurement. The appropriateness of the non-linear autoregression-based scheme has been tested for varying fault scenarios by considering wide variation in the fault attributes, non-linear loading, and noise level in the sensor measurement.

Length at maturity of two pelagic sharks (Isurus paucus and Carcharhinus longimanus) found off northern Cuba

Many shark species show regional structure and seasonal movements, complicating the gathering of life history data needed for fisheries management. Length-at-maturity is a relatively straightforward trait to quantify through the sampling of fisheries landings and is of great value for stock assessment. We report here on the median size at maturity for the longfin mako (Isurus paucus) and the oceanic whitetip shark (Carcharhinus longimanus), two important pelagic sharks caught by the small-scale longline fishery operating off northwest Cuba. We determined the maturity stages for 118 longfin makos and 84 oceanic whitetip sharks, based on the development of internal and external organs. A binary logistic regression was applied to estimate model coefficients necessary to describe proportion mature at size and determine L50 and the steepness of the model (φ). For the longfin mako, the parameters of logistic models for males were L50 = 215 cm TL (CI: 210–220 cm TL), φ = 4.46 (CI: 2.30–8.64) and for females L50 = 230 cm TL (CI: 220–239 cm TL), φ = 9.79 (CI: 5.15–18.62). For the oceanic whitetip shark, the parameters for males were L50 = 203 cm TL (CI: 186–220 cm TL), φ = 14.59 (CI: 6.74–31.58) and for females L50 = 199 cm TL (CI: 187–211 cm TL), φ = 7.48 (CI: 2.56–21.8). The parameters presented here contribute to stock assessment and management measures for global conservation of these transboundary, highly vulnerable species of pelagic sharks.

One-step correction strategy for BDS-2/BDS-3 satellite observation code bias and multipath delay

Global navigation satellite systems and positioning, navigation, and timing services, such as the newly developed BeiDou satellite system (BDS), require high-accuracy satellite observations. Satellite-induced code bias is present in BDS-2 satellite observations while it is negligibly present in BDS-3 satellite observations. The traditional method of mitigating BDS code bias involves two steps, first addressing multipath delay and then code bias, and does not obtain optimal results. A one-step strategy is therefore proposed to model and eliminate code bias and multipath delay considering BDS-2 and BDS-3 integrated processing. A combined least-square and autoregressive strategy is selected to estimate the model coefficients of code bias and construct the multipath delay with one solution. Moreover, inter-satellite correlations of BDS-2 and BDS-3 are extracted to improve the weight matrix in the estimation of model coefficients. To verify the proposed strategy, experiments are designed for eight schemes to analyze the coefficients and residuals of modelling code bias. Experimental results indicate that a more stable and accurate code bias model is acquired by introducing inter-satellite correlation; with the one-step strategy, the model residuals are reduced and a more optimal code bias model is output. Meanwhile, the single-frequency precise point positioning (PPP) and real-time PPP of BDS-2 and BDS-3 combined estimations are tested for different stations, frequencies, and code bias models. The results of final positions reveal that accuracy in the Up (U) direction is improved compared with that in the East (E) and North (N) directions; improvements are greater for B1I than for B3I. In general, the one-step strategy enhances the precision of single-frequency (B1I/B3I) PPP, especially in the U direction. However, the effects on BDS ultra-rapid orbit determination are negligible. Equally, the double-frequency real-time PPP solution is improved, compared with adopting the traditional method, by 10.6–64.9%, 0.0–59.1%, and 12.6–67.2% for E, N, and U directions, respectively, through BDS-2/BDS-3 integrated processing. The proposed one-step strategy therefore outperforms the two-step strategy in terms of rapidly processing high-precision BDS observations.

Comprehensive Evaluation of a Sparse Dataset, Assessment and Selection of Competing Models

With tremendous associated economic and social costs of crashes, researchers have been trying not only to identify the factors affecting crashes, but also to estimate those coefficients in the most accurate ways. Estimating model coefficients without accounting for a correct distribution would result in biased and erroneous results. This risk especially holds true when modeling skewed equivalent property damage only (EPDO) crashes with a preponderance of zeroes. The distribution of EPDO is known for not being modeled with known distributions such as Poisson or negative binomial distributions. This issue is highlighted in particular for a mountainous state like Wyoming that has very low traffic levels and a severely high crash rate. In addition, we included barriers in the model that did not experience any crashes but did suffer from being under-designed by geometric architects, thereby adding to the number of zero count observations. Various models with different distributional characteristics were considered and compared in this study. Comparisons were not just made across models in terms of their goodness of fit, but the estimated coefficients were also compared to see the impact of considering the wrong distributional assumptions on model parameter estimates. As the objectives of this study are to implement the identified results for optimization purposes and locate hazardous locations that could host future crashes, the results highlight accurate model estimations and the consequences of a failure to account for the right distributions. After conducting different goodness-of-fit measures, a hurdle model was proposed in this study to accommodate observations with zero crashes, and to account for a sparse distribution of EPDO crashes in the state of Wyoming. For the hurdle model, binary logistic regression was used to account for observations with zero crashes, while the negative binomial method was considered for non-zero observations. The findings of this study have direct implications on the allocation of limited funds for policymakers in Wyoming, as optimization could be conducted on the geometric characteristics of traffic barriers in future studies.

Rapid Assessment at Hospital Admission of Mortality Risk From COVID-19: The Role of Functional Status

ObjectiveTo evaluate the role of functional status along with other used clinical factors on the occurrence of death in patients hospitalized with COVID-19. DesignProspective cohort study. SettingPublic university hospital (Madrid). Participants and MethodsA total of 375 consecutive patients with COVID-19 infection, admitted to a Public University Hospital (Madrid) between March 1 and March 31, 2020, were included in the Prospective Cohort study. Death was the main outcome. The main variable was disability in activities of daily living (ADL) assessed with the Barthel Index. Covariates included sex, age, severity index (Quick Sequential Organ Failure Assessment, qSOFA), polypharmacy (≥5 drugs in the month before admission), and comorbidity (≥3 diseases). Multivariable logistic regression was used to identify risk factors for adverse outcomes. Estimated model coefficients served to calculate the expected probability of death for a selected combination of 5 variables: Barthel Index, sex, age, comorbidities, and severity index (qSOFA). ResultsMean age was 66 years (standard deviation 15.33), and there were 207 (55%) men. Seventy-four patients died (19.8%). Mortality was associated with low Barthel Index (odds ratio per 5-point decrease 1.11, 95% confidence interval 1.03-1.20), male sex (0.23, 0.11-0.47), age (1.07, 1.03-1.10), and comorbidity (2.15, 1.08-4.30) but not with qSOFA (1.29, 0.87-1.93) or polypharmacy (1.54, 0.77-3.08). Calculated mortality risk ranged from 0 to 0.78. Conclusions and ImplicationsFunctional status predicts death in hospitalized patients with COVID-19. Combination of 5 variables allows to predict individual probability of death. These findings provide useful information for the decision-making process and management of patients.

A Clinical Prediction Model for Normalization of Pulmonary Vascular Resistance After Left Ventricular Assist Device Implantation

Background Patients with heart failure with reduced ejection fraction (HFrEF) and elevated pulmonary vascular resistance (PVR) are at risk increased for poor outcomes and may be ineligible for heart transplantation (HT). Left ventricular assist device (LVAD) therapy can lower PVR for certain patients, but it is difficult to identify these patients preoperatively. Methods Using the INTERMACS registry, we developed a clinical prediction model to predict the likelihood of PVR normalization (≤ 3 WU) or HT within 1 year of LVAD implant among patients with baseline PVR > 3 WU. Backwards stepwise selection was used to select variables from 14 candidate predictors and logistic regression was used to estimate model coefficients. Model performance was summarized using measures of discrimination (c-statistic) and calibration (Harrel's EAVG). Internal validation was performed using bootstrap resampling over 200 iterations to generate an average calibration slope, which was used to uniformly shrink all model coefficients to generate the final model. Results There were 2209 patients available for analysis. Median preoperative PVR was 4.26 WU (IQR 3.54-5.54 WU) and the median time from LVAD implant to postoperative PVR assessment was 3 months. 1596 patients (72%) experienced PVR normalization or were transplanted within 1 year. 412 (19%) patients died and 201 (9%) were alive with persistently elevated PVR at 1 year. Final model coefficients are shown in the Table. Important predictors included older age (OR 0.72 per 10 years), severe baseline mitral regurgitation (MR) (OR 2.49 vs no MR), and increased baseline pulmonary capillary wedge pressure (PCWP) (OR 1.24 per 5 mmHg increase). The model c-statistic was 0.66 and EAVG was 0.004. Conclusion Among patients with elevated PVR, preoperative patient characteristics such as severe MR and elevated PCWP are able to predict the likelihood of PVR normalization or HT within 1 year of LVAD implantation. These findings can help to inform decision making and frame patient expectations regarding anticipated improvements in pulmonary vascular disease following LVAD surgery. Future studies are needed to investigate potential management strategies to increase the likelihood of PVR normalization with LVAD support.

Visibility Modeling and Prediction for Free Space Optical Communication Systems for South Africa

Due to the cost and complexity in the measurement of Free Space Optical (FSO) visibility, this paper presents regression models based on meteorological factors to reliably estimate atmospheric visibility. The meteorological factors used are relative humidity, temperature, fractional sunshine, atmospheric pressure and wind speed for Cape Town, South Africa. Initially, Simple Linear Regression (SLR) models are developed and presented. To improve the performance of the regression, the SLR model is extended to a Multiple Linear Regression model (MLR) where three of the meteorological factors are taken into consideration simultaneously. It was found that by implementing MLR, the model performance improves considerably. However, it was also found that the model had effects of multicollinearity due to some of the predictor variables being highly correlated. To mitigate the effects of multicollinearity, two approaches are proposed, 1) removing the problematic terms from the regression model and 2) introducing interaction terms. Both approaches are seen to have little impact on the overall performance of the MLR model while the estimated model coefficients are significant at 5% significant level. In general, it is found through application of standard statistical tests that both SLR and MLR models can be used adequately to determine visibility at a location.

Sparse Logistic Regression: Comparison of Regularization and Bayesian Implementations

In knowledge-based systems, besides obtaining good output prediction accuracy, it is crucial to understand the subset of input variables that have most influence on the output, with the goal of gaining deeper insight into the underlying process. These requirements call for logistic model estimation techniques that provide a sparse solution, i.e., where coefficients associated with non-important variables are set to zero. In this work we compare the performance of two methods: the first one is based on the well known Least Absolute Shrinkage and Selection Operator (LASSO) which involves regularization with an ℓ 1 norm; the second one is the Relevance Vector Machine (RVM) which is based on a Bayesian implementation of the linear logistic model. The two methods are extensively compared in this paper, on real and simulated datasets. Results show that, in general, the two approaches are comparable in terms of prediction performance. RVM outperforms the LASSO both in term of structure recovery (estimation of the correct non-zero model coefficients) and prediction accuracy when the dimensionality of the data tends to increase. However, LASSO shows comparable performance to RVM when the dimensionality of the data is much higher than number of samples that is p > > n .

Penalized and ridge-type shrinkage estimators in Poisson regression model

The paper considers the problem of estimation of the regression coefficients in a Poisson regression model under multicollinearity situation. We propose non-penalty Stein-type shrinkage ridge estimation approach when it is conjectured that some prior information is available in the form of potential linear restrictions on the coefficients. We establish the asymptotic distributional biases and risks of the proposed estimators and investigate their relative performance with respect to the unrestricted ridge estimator. For comparison sake, we consider the two penalty estimators, namely, least absolute shrinkage and selection operator and Elastic-Net estimators and compare numerically their relative performance with the other listed estimators. Monte-Carlo simulation experiment is conducted to evaluate the performance of each estimator in terms of the simulated relative efficiency. The results show that the shrinkage ridge estimators perform better than the penalty estimators in certain parts of the parameter space. Finally, a real data example is illustrated to evaluate of the proposed methods.

Tracer kinetic models as temporal constraints during brain tumor DCE-MRI reconstruction.

PurposeTo apply tracer kinetic models as temporal constraints during reconstruction of under‐sampled brain tumor dynamic contrast enhanced (DCE) magnetic resonance imaging (MRI).MethodsA library of concentration vs time profiles is simulated for a range of physiological kinetic parameters. The library is reduced to a dictionary of temporal bases, where each profile is approximated by a sparse linear combination of the bases. Image reconstruction is formulated as estimation of concentration profiles and sparse model coefficients with a fixed sparsity level. Simulations are performed to evaluate modeling error, and error statistics in kinetic parameter estimation in presence of noise. Retrospective under‐sampling experiments are performed on a brain tumor DCE digital reference object (DRO), and 12 brain tumor in‐vivo 3T datasets. The performances of the proposed under‐sampled reconstruction scheme and an existing compressed sensing‐based temporal finite‐difference (tFD) under‐sampled reconstruction were compared against the fully sampled inverse Fourier Transform‐based reconstruction.ResultsSimulations demonstrate that sparsity levels of 2 and 3 model the library profiles from the Patlak and extended Tofts‐Kety (ETK) models, respectively. Noise sensitivity analysis showed equivalent kinetic parameter estimation error statistics from noisy concentration profiles, and model approximated profiles. DRO‐based experiments showed good fidelity in recovery of kinetic maps from 20‐fold under‐sampled data. In‐vivo experiments demonstrated reduced bias and uncertainty in kinetic mapping with the proposed approach compared to tFD at under‐sampled reduction factors >= 20.ConclusionsTracer kinetic models can be applied as temporal constraints during brain tumor DCE‐MRI reconstruction. The proposed under‐sampled scheme resulted in model parameter estimates less biased with respect to conventional fully sampled DCE MRI reconstructions and parameter estimation. The approach is flexible, can use nonlinear kinetic models, and does not require tuning of regularization parameters.

SDA: a semi-parametric differential abundance analysis method for metabolomics and proteomics data

BackgroundIdentifying differentially abundant features between different experimental groups is a common goal for many metabolomics and proteomics studies. However, analyzing data from mass spectrometry (MS) is difficult because the data may not be normally distributed and there is often a large fraction of zero values. Although several statistical methods have been proposed, they either require the data normality assumption or are inefficient.ResultsWe propose a new semi-parametric differential abundance analysis (SDA) method for metabolomics and proteomics data from MS. The method considers a two-part model, a logistic regression for the zero proportion and a semi-parametric log-linear model for the possibly non-normally distributed non-zero values, to characterize data from each feature. A kernel-smoothed likelihood method is developed to estimate model coefficients and a likelihood ratio test is constructed for differential abundant analysis. The method has been implemented into an R package, SDAMS, which is available at https://www.bioconductor.org/packages/release/bioc/html/SDAMS.html.ConclusionBy introducing the two-part semi-parametric model, SDA is able to handle both non-normally distributed data and large fraction of zero values in a MS dataset. It also allows for adjustment of covariates. Simulations and real data analyses demonstrate that SDA outperforms existing methods.

Spatial Auto-Regressive Analysis of Correlation in 3-D PET With Application to Model-Based Simulation of Data.

When a scanner is installed and begins to be used operationally, its actual performance may deviate somewhat from the predictions made at the design stage. Thus it is recommended that routine quality assurance (QA) measurements be used to provide an operational understanding of scanning properties. While QA data are primarily used to evaluate sensitivity and bias patterns, there is a possibility to also make use of such data sets for a more refined understanding of the 3-D scanning properties. Building on some recent work on analysis of the distributional characteristics of iteratively reconstructed PET data, we construct an auto-regression model for analysis of the 3-D spatial auto-covariance structure of iteratively reconstructed data, after normalization. Appropriate likelihood-based statistical techniques for estimation of the auto-regression model coefficients are described. The fitted model leads to a simple process for approximate simulation of scanner performance-one that is readily implemented in an R script. The analysis provides a practical mechanism for evaluating the operational error characteristics of iteratively reconstructed PET images. Simulation studies are used for validation. The approach is illustrated on QA data from an operational clinical scanner and numerical phantom data. We also demonstrate the potential for use of these techniques, as a form of model-based bootstrapping, to provide assessments of measurement uncertainties in variables derived from clinical FDG-PET scans. This is illustrated using data from a clinical scan in a lung cancer patient, after a 3-minute acquisition has been re-binned into three consecutive 1-minute time-frames. An uncertainty measure for the tumor SUVmax value is obtained. The methodology is seen to be practical and could be a useful support for quantitative decision making based on PET data.

Some interesting behaviors of good lattice point sets

Good lattice point (GLP) sets are sets of points that are uniformly distributed over the domain of interest and thus have good space-filling property. GLP sets are an important kind of points for many applications, such as multidimensional quadrature, simulation, computer experiments, quasi-Monte Carlo techniques and design of experiments. It is a significant issue to study the behaviors of GLP sets. For instance, most real-life applications require that the columns of a GLP set are independent, i.e., the GLP set has full rank. If one or more columns are linearly dependent, there will be more confounding among the main effects and interactions in statistical models and also we can not find the least squares estimation of regression coefficients in the linear regression model. The problem of finding the rank of the GLP set remained unsolved since its inception in 1959. It is desirable to put the first stone for solving this significant problem. Through theoretical justifications, this paper gives some interesting behaviors of the generating vector of any GLP set and shows the independence among its rows and columns by presenting some analytic linkages among these rows and columns. The results of this paper are used not only as a benchmark for constructing full rank GLP sets, but also in various applications of GLP sets.

DeconvSeq: deconvolution of cell mixture distribution in sequencing data.

Although single-cell sequencing is becoming more widely available, many tissue samples such as intracranial aneurysms are both fibrous and minute, and therefore not easily dissociated into single cells. To account for the cell type heterogeneity in such tissues therefore requires a computational method. We present a computational deconvolution method, deconvSeq, for sequencing data (RNA and bisulfite) obtained from bulk tissue. This method can also be applied to single-cell RNA sequencing data. DeconvSeq utilizes a generalized linear model to model effects of tissue type on feature quantification, which is specific to the data structure of the sequencing type used. Estimated model coefficients can then be used to predict the cell type mixture within a tissue. Predicted cell type mixtures were validated against actual cell counts in whole blood samples. Using this method, we obtained a mean correlation of 0.998 (95% CI 0.995-0.999) from the RNA sequencing data of 35 whole blood samples and 0.95 (95% CI 0.91-0.98) from the reduced representation bisulfite sequencing data from 35 whole blood samples. Using symmetric balances to obtain the correlation between compositional parts, we found that the lowest correlation occurred for monocytes for both RNA and bisulfite sequencing. Comparison with other methods of decomposition such as deconRNAseq, CIBERSORT, MuSiC and EpiDISH showed that deconvSeq is able to achieve good prediction using mean correlation with far fewer genes or CpG sites in the signature set. Software implementing deconvSeq is available at https://github.com/rosedu1/deconvSeq. Supplementary data are available at Bioinformatics online.

Aggregating the conceptualization of movement data better captures real world and simulated animal–environment relationships

ABSTRACT Habitat selection analysis is a widely applied statistical framework used in spatial ecology. Many of the methods used to generate movement and couple it with the environment are strongly integrated within GIScience. The choice of movement conceptualisation and environmental space can potentially have long-lasting implications on the spatial statistics used to infer movement–environment relationships. The aim of this study was to explore how systematically altering the conceptualisation of movement, environmental space and temporal resolution affects the results of habitat selection analyses using both real-world case studies and a virtual ecologist approach. Model performance and coefficient estimates did not differ between the finest conceptualisations of movement (e.g. vector and move), while substantial differences were found for the more aggregated representations (e.g. segment and area). Only segments modelled the expected movement–environment relationship with increasing linear feature resistance in the virtual ecologist approach and altering the temporal resolution identified inversions in the movement–environment relationship for vectors and moves. The results suggest that spatial statistics employed to investigate movement–environment relationships should advance beyond conceptualising movement as the (relatively) static conceptualisation of vectors and moves and replace these with (more) dynamic aggregations of longer-lasting movement processes such as segments and areal representations.

Multivariate copula temporal modeling of intersection crash consequence metrics: A joint estimation of injury severity, crash type, vehicle damage and driver error

This study employs a copula-based multivariate temporal ordered probit model to simultaneously estimate the four common intersection crash consequence metrics – driver error, crash type, vehicle damage and injury severity – by accounting for potential correlations due to common observed and unobserved factors, while also accommodating the temporal instability of model estimates over time. To this end, a comprehensive literature review of relevant studies was conducted; four different copula model specifications including Frank, Clayton, Joe and Gumbel were estimated to identify the dominant factors contributing to each crash consequence indicator; the temporal effects on model estimates were investigated; the elasticity effects of the independent variables with regard to all four crash consequence indicators were measured to express the magnitude of the effects of an independent variable on the probability change for each level of four indicators; and specific countermeasures were recommended for each of the contributing factors to improve the intersection safety.The model goodness-of-fit illustrates that the Joe copula model with the parameterized copula parameters outperforms the other models, which verifies that the injury severity, crash type, vehicle damage and driver error are significantly correlated due to common observed and unobserved factors and, accounting for their correlations, can lead to more accurate model estimation results. The parameterization of the copula function indicates that their correlation varies among different crashes, including crashes that occurred at stop-controlled intersections, four-leg intersections and crashes which involved drivers younger than 25. The model coefficient estimates indicate that the driver’s age, driving under the influence of drugs and alcohol, intersection geometry and control types, and adverse weather and light conditions are the most critical factors contributing to severe crash consequences. The coefficient estimates of four-leg intersections, yield and stop-controlled intersections and adverse weather conditions varied over time, which indicates that the model estimation of crash data may not be stable over time and should be accommodated in crash prediction analysis. In the end, relevant countermeasures corresponding to law enforcement and intersection infrastructure design are recommended to all of the contributing factors identified by the model. It is anticipated that this study can shed light on selecting valid statistical models for crash data analysis, identifying intersection safety issues, and helping develop effective countermeasures to improve intersection safety.

Multi-study inference of regulatory networks for more accurate models of gene regulation.

Gene regulatory networks are composed of sub-networks that are often shared across biological processes, cell-types, and organisms. Leveraging multiple sources of information, such as publicly available gene expression datasets, could therefore be helpful when learning a network of interest. Integrating data across different studies, however, raises numerous technical concerns. Hence, a common approach in network inference, and broadly in genomics research, is to separately learn models from each dataset and combine the results. Individual models, however, often suffer from under-sampling, poor generalization and limited network recovery. In this study, we explore previous integration strategies, such as batch-correction and model ensembles, and introduce a new multitask learning approach for joint network inference across several datasets. Our method initially estimates the activities of transcription factors, and subsequently, infers the relevant network topology. As regulatory interactions are context-dependent, we estimate model coefficients as a combination of both dataset-specific and conserved components. In addition, adaptive penalties may be used to favor models that include interactions derived from multiple sources of prior knowledge including orthogonal genomics experiments. We evaluate generalization and network recovery using examples from Bacillus subtilis and Saccharomyces cerevisiae, and show that sharing information across models improves network reconstruction. Finally, we demonstrate robustness to both false positives in the prior information and heterogeneity among datasets.

Bayesian geographically weighted regression and its application for local modeling of relationships between tree variables

Geographically weighted regression (GWR) has become popular in recent years to deal with spatial autocorrelation and heterogeneity in forestry and ecological data. However, researchers have realized that GWR has some limitations, such as correlated model coefficients across study areas, strong influence of outliers, weak data problem, etc. In this study, we applied Bayesian geographically weighted regression (BGWR) and a robust BGWR (rBGWR) to model the relationship between tree crown and diameter using observed tree data and simulated data to investigate model fitting and performance in order to overcome some limitations of GWR. Our results indicated that, for observed tree data, the rBGWR estimated tree crown more accurate than both BGWR and GWR. For the simulated data, 74.1% of the estimated slope coefficients by rBGWR and 73.4% of the estimated slope coefficients by BGWR were not significantly different (α = 0.05) from the corresponding simulated slope coefficients. The estimation of model coefficients by rBGWR was not sensitive to outliers, but the coefficient estimation by GWR was strongly affected by those outliers. The majority of the coefficient estimates by rBGWR and BGWR for weak observations were not significantly (α = 0.05) different from the simulated coefficients. Therefore, BGWR (including rBGWR) may be a better alternative to overcome some limitations of GWR. In addition, both BGWR and rBGWR were more powerful than GWR to detect the spatial areas with non-constant variance or spatial outliers.