Empirical Bayesian Model Research Articles

SU106 - DIFFERENTIAL DISTRIBUTION OF SCHIZOPHRENIA RISK GENES WITHIN GENE CO-EXPRESSION NETWORKS CONSTRUCTED FROM RNA-SEQ DATA (POSTMORTEM DLPFC) OF AFFECTED AND UNAFFECTED INDIVIDUALS

Schizophrenia polygenic risk is plausibly manifested by complex transcriptional dysregulation in the brain, involving networks of co-expressed and functionally related genes. 1. Samples: DLPFC gene expression (RNA-Seq) abundance was quantified (RPKM) in postmortem human brains from the LIBD Postmortem Human Brain Repository (90 controls- CTRL, 76 schizophrenia- SCZ, Caucasians, age: 16–80, RIN≥7). Common Mind Consortium (CMC) RNA-Seq postmortem DLPFC (125 CTRL, 109 SCZ, Caucasians, age, RIN as above) was used for replication of LIBD co-expression networks. 2. Expression measures/ quality control: only genes with median RPKM≥0.1 (N=23132 genes; LIBD samples) were used. Expression data was normalized by log2 transformation and adjusted for RNA quality measures (RIN, PMI, total gene assignment, proportion of mitochondrial RNA) by empirical Bayesian models. 3. Data analysis: WGCNA was applied to the adjusted expression data to construct co-expression networks, separately for LIBD- CTRL and LIBD- SCZ. Modules of co-expressed genes were detected by dynamic tree cutting method and summarized as “module eigengenes” for further analyses. 4. Modules of LIBD (CTRL; SCZ) were tested for enrichment in PGC2 genes within the 108 loci associated with the schizophrenia risk in the latest GWAS. 5. Composite module preservation statistics- median rank and Zsummary was used for assessing the preservation of LIBD SCZ modules in LIBD CTRL network and for replication of LIBD co-expression modules in the CMC independent data sets. Gene enrichment analyses for testing modules’ enrichment in Gene Ontology (GO) biological processes (BP) was performed with AmiGO/ PANTHER. 17 co-expression modules were identified in each LIBD group. Module preservation analysis showed a good preservation of co-expression modules between LIBD CTRL and SCZ networks, with all but one SCZ module showing evidence for strong preservation (Zsummary≥10). PGC2 genes were overrepresented in two CTRL modules (p=0.001973, p=0.028297) enriched for GO-BP such as translation, ATP synthesis, oxidative phosphorylation, homophilic cell adhesion via plasma membrane adhesion molecules, chemical synaptic transmission, axonogenesis, and in one SCZ module (p=0.000519) enriched for RNA processing, nervous system development and modulation of synaptic transmission. 58 PGC2 genes overlapped with the SCZ module (e.g., DRD2, FURIN, TSNARE1), whereas 48 + 17 PGC2 genes were overrepresented in the two CTRL modules (e.g., CACNA1C, CHRM4, CACNA1I, TCF20). Of note, only 9 overrepresented PGC2 genes were shared by CTRL and SCZ: AMBRA1, ANKRD63, HSPA9, LRP1, PRR12, PTPRF, RERE, TOM1L2, ZDHHC5. Replication of co-expression LIBD networks in CMC data set showed evidence of weak to strong preservation (Zsummary>2, respectively ≥10) for 16/ 17 modules for each LIBD group. Our results illustrate the complexity of SCZ risk distribution within the DLPFC co-expression networks in postmortem brains of affected and unaffected individuals. Importantly, not the same PGC2 genes are members of co-expression modules in CTRL and SCZ, nor do they aggregate in similar modules. These observations may be explained by differential transcriptional regulation due to pleiotropy, epistasis, epigenetic dysregulation, or even hidden RNA quality artifacts. Future studies are necessary to elucidate the participation of SCZ risk genes in brain co-expression networks.

Geographical Variation in Oral and Oropharynx Cancer Mortality in Brazil: A Bayesian Approach.

Recent studies have shown a high number of deaths from oral and oropharyngeal cancer worldwide, Brazil included. For this study, the deaths data (ICD-10, chapter II, categories C00 to C14) was obtained from Mortality Information System (SIM) and standardized by gender and population for each of the 554 Microregions of Brazil. The raw mortality rates were adopted as the standard and compared to the application of smoothing by the Bayesian model. In order to describe the geographical pattern of the occurrence of oral cancer, thematic maps were constructed, based on the distributions of mortality rates for Microregions and gender. Results: There were 7882 deaths registered due to oral and oropharyngeal cancer in Brazil, of which 6291 (79.81%) were male and 1591 (20.19%) female. The Empirical Bayesian Model presented greater scattering with mosaic appearance throughout the country, depicting high rates in Southeast and South regions interpolated with geographic voids of low rates in Midwest and North regions. For males, it was possible to identify expressive clusters in the Southeast and South regions. Conclusion: The Empirical Bayesian Model allowed an alternative interpretation of the oral and oropharynx cancer mortality mapping in Brazil.

Trend and spatial analysis of prostate cancer mortality in the state of Sergipe, Brazil.

This is an ecological study with exploratory analysis of spatial and temporal data based on mortality data with respect to prostate cancer obtained from the Mortality Information System concerning residents of the state of Sergipe, Brazil between 2000 and 2015. The analysis of temporal trends was performed using the Joinpoint Regression Program through Poisson regression. Spatial analysis was performed using the empirical Bayesian model, Kernel analysis, Global Moran and Local indices. There were 1,986 deaths due to prostate cancer, most of which occurring after 60 years of age. An increasing, non-constant but significant trend in mortality rates was noted. The kernel density estimator showed hotspot densities of the highest rates of prostate cancer mortality in the north-eastern and central regions of the state. High-risk clusters were identified for prostate cancer mortality (I = 0.55, P<0.01). There was an increase in prostate cancer mortality rates and a heterogeneous geographic distribution of risk areas, with high-risk priority areas identified in certain regions of the state. These priority areas include the municipalities located in the Northeast (Amparo do São Francisco, Aquidabã, Canhoba, Cedro de São João and Telha), the West (Frei Paulo and Pedra Mole) and the south-western region of the state (Poço Verde and Simão Dias).

Spatial study of leprosy in Bahia, Brazil, 2001-2012: an approach based on the local empirical Bayesian model.

to compare the temporal evolution and spatial distribution of epidemiological indicators of leprosy, both crude and also corrected using the empirical Bayesian model, Bahia, Brazil, 2001-2012. this was an ecological study using data from the Notifiable Diseases Information System; all 417 municipalities in Bahia were included and the following indicators per 100,000 inhabitants were analyzed - detection rate of new cases in the general population, in those <15 years old, and in those with grade 2 physical disability -; the local empirical Bayesian model was used to smoothen the indicators, and Student's t-test was used to compare means. indicators estimated by the model were higher than crude indicators; estimated detection rates in the general population and in those <15 years old were higher than crude rates in 253 (60.7%) and 209 (50.1%) municipalities, respectively; areas of greatest risk were concentrated in the northwestern and southern regions of the state. spatial distribution of the disease was heterogeneous and there was possible underreporting of cases.

Poisson-generalized gamma empirical Bayes model for disease mapping

In spatial disease mapping, the use of Bayesian models of estimation technique is becoming popular for smoothing relative risks estimates for disease mapping. The most common Bayesian conjugate model for disease mapping is the Poisson-Gamma Model (PG). To explore further the activity of smoothing of relative risk estimates for Bayesian disease mapping, this study focused on the use of generalized gamma distribution as conjugate priors with respect to Poisson likelihood. Two new empirical Bayesian (EB) models are developed; these include Poisson-Generalized Gamma model (PGG) and modified Poisson-Generalized Gamma model (MPGG). The model simulation results indicated that PGG and MPGG models are more likely to handle dispersion in zero-deflated data, contaminated data and zero-inflated data for small and large sample data. Hence, the new EB models are highly competitive to improve the efficiency of relative risk estimation for disease mapping.Keywords: Disease Mapping, Empirical Bayes, Generalized Gamma, Dispersion, Poisson

Bayesian model and spatial analysis of oral and oropharynx cancer mortality in Minas Gerais, Brazil.

The objective of this study was to determine of oral and oropharynx cancer mortality rate and the results were analyzed by applying the Spatial Analysis of Empirical Bayesian Model. To this end, we used the information contained in the International Classification of Diseases (ICD-10), Chapter II, Category C00 to C14 and Brazilian Mortality Information System (SIM) of Minas Gerais State. Descriptive statistics were observed and the gross rate of mortality was calculated for each municipality. Then Empirical Bayesian estimators were applied. The results showed that, in 2012, in the state of Minas Gerais, were registered 769 deaths of patients with cancer of oral and oropharynx, with 607 (78.96%) men and 162 (21.04%) women. There was a wide variation in spatial distribution of crude mortality rate and were identified agglomeration in the South, Central and North more accurately by Bayesian Estimator Global and Local Model. Through Bayesian models was possible to map the spatial clustering of deaths from oral cancer more accurately, and with the application of the method of spatial epidemiology, it was possible to obtain more accurate results and provide subsidies to reduce the number of deaths from this type of cancer.

Prognostic Factors of Breast Cancer Survival in the Islamic Republic of Iran: an Additive Empirical Bayesian Approach.

Breast cancer is a public health challenge in the Islamic Republic of Iran. Relatively few studies have dealt with large dataset and advanced statistical methods. Therefore, we aimed to investigate the role of prognostic factors on breast cancer survival using Additive Empirical Bayesian model with large data set. Breast cancer data set included 1574 women diagnosed with breast cancer from 2002 to 2012 that registered from Cancer Registry in Fars Province, Islamic Republic of Iran. Overall survival rates at 2, 3, 5 and 10 years were 0.98, 0.94, 0.87 and 0.76, respectively. Five years survival at stages 1, 2 and 3 were 0.94, 0.92 and 0.74, respectively. The younger patients with characteristics such as zero involved nodes, negative progesterone receptor, free skin and good prognostic level had a higher survival chance than others. The 5-year survival probability by stage in Fars Province was nearly the same as that reported by the American Cancer Society. The Nottingham prognostic index (NPI) related to nodal status, tumour size and nuclear grade was the main indicator of breast cancer mortality.

High-Throughput and Sensitive Quantification of Circulating Tumor DNA by Microfluidic-Based Multiplex PCR and Next-Generation Sequencing

Circulating tumor DNA (ctDNA) has potential to serve as a biomarker for noninvasive monitoring of treatment response and disease progression. However, broad clinical applicability of ctDNA has been limited by the low sensitivity, throughput, and patient coverage offered by existing ctDNA detection methods. Herein, we report the adaptation and characterization of the microfluidics multiplex PCR sequencing technology for high-throughput and sensitive quantitation of ctDNA. A multiplex PCR preamplification step was developed and incorporated into the microfluidics multiplex PCR sequencing work flow to enable low-input ctDNA analysis with enhanced sensitivity. An empirical bayesian model was developed to characterize both position and substitution-associated system errors specific to this platform and provided a tailored approach to greatly enhance the confidence and accuracy of variant calling for ctDNA analysis. Clinical validation of this platform for ctDNA mutation detection demonstrated an overall sensitivity of 92% and specificity of 100% when using mutation calls in the matched tumor tissues as a benchmark. Finally, we established an early proof of concept of clinical utility of this ctDNA work flow for monitoring disease progression using clinical trial samples. Our novel ctDNA work flow provides a high-throughput and sensitive platform that can be implemented in clinical trials for mutation detection and disease monitoring from plasma ctDNA.

Statistics of Earthquake Activity: Models and Methods for Earthquake Predictability Studies

Statistical methods and various models in time-space-magnitude parameter space of earthquakes are being developed to analyze seismic activity based on earthquake hypocenter catalogs that are routinely accumulated. Considering complex geophysical environments and uncertainties, we seek proper stochastic modeling that depends on the history of earthquake occurrences and relevant geophysical information for describing and forecasting earthquake activity. Also, we need empirical Bayesian models with many parameters in order to describe nonstationary or nonhomogeneous seismic activity. This review is concerned with earthquake predictability research aimed at realizing practical operational forecasting. In particular, uncertainty lies in identifying whether abnormal phenomena are precursors to large earthquakes. The predictability of such models can be examined by certain statistical criteria.

A comparison of machine learning and Bayesian modelling for molecular serotyping

BackgroundStreptococcus pneumoniae is a human pathogen that is a major cause of infant mortality. Identifying the pneumococcal serotype is an important step in monitoring the impact of vaccines used to protect against disease. Genomic microarrays provide an effective method for molecular serotyping. Previously we developed an empirical Bayesian model for the classification of serotypes from a molecular serotyping array. With only few samples available, a model driven approach was the only option. In the meanwhile, several thousand samples have been made available to us, providing an opportunity to investigate serotype classification by machine learning methods, which could complement the Bayesian model.ResultsWe compare the performance of the original Bayesian model with two machine learning algorithms: Gradient Boosting Machines and Random Forests. We present our results as an example of a generic strategy whereby a preliminary probabilistic model is complemented or replaced by a machine learning classifier once enough data are available. Despite the availability of thousands of serotyping arrays, a problem encountered when applying machine learning methods is the lack of training data containing mixtures of serotypes; due to the large number of possible combinations. Most of the available training data comprises samples with only a single serotype. To overcome the lack of training data we implemented an iterative analysis, creating artificial training data of serotype mixtures by combining raw data from single serotype arrays.ConclusionsWith the enhanced training set the machine learning algorithms out perform the original Bayesian model. However, for serotypes currently lacking sufficient training data the best performing implementation was a combination of the results of the Bayesian Model and the Gradient Boosting Machine. As well as being an effective method for classifying biological data, machine learning can also be used as an efficient method for revealing subtle biological insights, which we illustrate with an example.

Abstract 2740: High-throughput and sensitive quantification of circulating tumor DNA

Abstract Circulating tumor DNA (ctDNA) has a great potential to serve as non-invasive biomarkers for monitoring disease progression and treatment response. However, its broad clinical applications have been hindered by sensitivity, throughput and complexity of existing ctDNA detection methods. Here we report an adaptation of a multiplex PCR pre-amplification (PreAmp) into the microfluidic multiplex PCR (MMP-Seq) technology for high-throughput and sensitive quantitation of ctDNA. We have demonstrated that PreAmp MMP-Seq enabled sensitive (down to 0.6% allele frequency), reproducible (R2=0.97), accurate and quantitative mutation detection from as low as 1.6 ng DNA input. An Empirical Bayesian Model has been established to estimate background error rates for every base of amplicons in the ctDNA panel using data from 43 normal plasmas. 0.1-0.3% of median error rates were observed across all positions. 149 longitudinal plasma samples from 12 ovarian and 10 pancreatic cancer patients of PhI DMUC5754A anti-MUC16 ADC trials were then profiled using PreAmp MMP-Seq. 79% (11/14) driver mutations found in tissue samples were also identified in matching baseline plasma samples at frequencies well above estimated error rates. Moreover, we found that changes of corresponding ctDNA concentration in post-treatment plasma samples were correlated with changes in tumor burdens and circulating protein marker levels. In conclusion, the MMP-seq PreAmp workflow provides an inexpensive, reproducible and flexible solution for mutation detection in plasma DNA samples. Citation Format: Yinghui Guan, Oleg Mayba, Shan Lu, Thomas Sandmann, Younjeong Choi, Walter Darbonne, Vincent Leveque, Lisa Ryner, Eric Humke, Nga W. Tam, Sundari Sujathasarma, Anna Cheung, Richard Bourgon, Yulei Wang. High-throughput and sensitive quantification of circulating tumor DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 2740. doi:10.1158/1538-7445.AM2017-2740

Hierarchical Bayesian fatigue data analysis

The problem minimizing the number of specimens required for fatigue data analysis is considered in this research. Assuming unknown hyperparameters described via prior distributions, a hierarchical Bayesian model with accumulated prior information was proposed to deal with this issue. One of the main advantages of hierarchical Bayesian model over the empirical Bayesian model is that the prior distributions with hierarchical structure can incorporate structural prior and subjective prior simultaneously. The probabilistic stress-cycle (P-S-N) curves are generated from the predictive distributions, involving both the randomness of parameters and the scatter of observations, and calculated by an identical hierarchical structure. The numerical calculation is done via the Gibbs sampler, which makes the whole process simple and intuitive.

Impact of Distribution Type in Bayes Probability Flood Forecasting

Bayesian forecasting system (BFS) is widely applied to the hydrological forecast. Hydrological forecast processor (HUP), a key part of the Bayesian probability prediction, is conducted at the assumption that the rainfall is certain, which can simultaneously quantify the uncertainty of hydrological model and parameter. In the HUP, the runoff is usually assumed to obey Logweibull distribution or Normal distribution. However, Distribution type of the runoff is not certain at different areas, and there are few distribution types of HUP in existence. So common distribution types are needed to develop the HUP to provide an effective forecast result. In this paper, Nonparametric kernel density estimation, Pearson III and Empirical distribution were introduced as the prior distribution of HUP to eliminate the parameter uncertainty of probability density function. Also, the five distributions were compared in this study to get the diversity of distribution types and search the best appropriate distribution type. The 52 floods during 2004a-2014a of ZheXi basin are employed to calibrate and validate the different distribution types of BFS. The results show that the LogWeibull and Empirical Bayesian probabilistic model has the best performance on average results compared with the other four distribution models. Meanwhile the other distribution types proposed in this study have the similar ability on interval width and the containing rate of probability forecasting results. This demonstrates that more new distributions are required to make BFS more robust.

Occurrence and distribution of fluoride in the groundwater of the Tamiraparani River basin, South India: a geostatistical modeling approach

Fluoride in drinking water has both beneficial and detrimental effects on public health, and a narrow range between .6 and 1.5 mg/L is optimal for consumption. However, natural groundwater sources exceed these guidelines affecting the entire population. This study aims to assess the distribution and controlling factors of fluoride concentration in the Tamiraparani River basin, South India. A total of 124 groundwater samples were analyzed for their fluoride content and other hydrogeochemical parameters. The fluoride concentration in the study area varied from .01 to 1.67 mg/L, and the highest concentrations were measured in the northern and central parts of the study area, which is underlain by charnockites and hornblende biotite gneiss. The sampling indicated (as per the Bureau of Indian Standards) that 53.9% of the area has fluoride concentrations below levels that are protective of teeth from dental caries (<.6 mg/L). .1% of the area is considered to be at risk of dental fluorosis, and the remaining 46% of the area is considered to have fluoride levels at desirable to permissible limit in groundwater. The groundwater in the study area belongs to Ca–Mg–Cl–SO4 and Ca–Mg–HCO3 types. A positive correlation between fluoride and TDS, Na+, K+ and HCO3 − indicates its geogenic origin, and positive loading between pH and fluoride shows that alkaline environment enhances the dissolution of fluoride-bearing minerals into the groundwater. An empirical Bayesian kriging model was applied to interpolate the fluoride concentration in the study area. This geostatistical model is found to be better than other kriging methods, and it yielded an average standard error of .332 and root-mean-square standardized value of .986.

Exploratory Analysis of Dengue Fever Niche Variables within the Río Magdalena Watershed

Previous research on Dengue Fever have involved laboratory tests or study areas with less diverse temperature and elevation ranges than is found in Colombia; therefore, preliminary research was needed to identify location specific attributes of Dengue Fever transmission. Environmental variables derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) and Tropical Rainfall Measuring Mission (TRMM) satellites were combined with population variables to be statistically compared against reported cases of Dengue Fever in the Río Magdalena watershed, Colombia. Three-factor analysis models were investigated to analyze variable patterns, including a population, population density, and empirical Bayesian estimation model. Results identified varying levels of Dengue Fever transmission risk, and environmental characteristics which support, and advance, the research literature. Multiple temperature metrics, elevation, and vegetation composition were among the more contributory variables found to identify future potential outbreak locations.

Methodology for Evaluating Influences of Anthropogenic Factors on Temperature Formation of Summer Seasons Considering their Randomness -Verification Through the Hierarchical Bayesian Model-

A new regression model that considers random nature is proposed for describing summer air temperatures in urban areas through the concept of an empirical hierarchical Bayesian model. In this model, the regression coefficients of explanatory variables are regarded as having random effects because of their locations, whereas ordinal regression models do not allow such randomness. Data of air temperature, the climatic situation, and geographically related factors in the Kinki area were collected and compiled by GIS, and then applied to both the new model and the standard regression models. The performances of the models were compared and their differences were discussed in relation to the geographic or climatic situations. It was established that the proposed model succeeded in accounting for the random nature and it predicted air temperature more accurately than the standard models based on the MCMC simulation. Furthermore, the confidence intervals of the values predicted by the new model became more rigid than for the ordinal regression models. This means that the new model is capable of presenting a more accurate picture without affecting the convenience of the ordinary regression models.

Low birth weight and air pollution in California: Which sources and components drive the risk?

IntroductionIntrauterine growth restriction has been associated with exposure to air pollution, but there is a need to clarify which sources and components are most likely responsible. This study investigated the associations between low birth weight (LBW, <2500g) in term born infants (≥37 gestational weeks) and air pollution by source and composition in California, over the period 2001–2008. MethodsComplementary exposure models were used: an empirical Bayesian kriging model for the interpolation of ambient pollutant measurements, a source-oriented chemical transport model (using California emission inventories) that estimated fine and ultrafine particulate matter (PM2.5 and PM0.1, respectively) mass concentrations (4km×4km) by source and composition, a line-source roadway dispersion model at fine resolution, and traffic index estimates. Birth weight was obtained from California birth certificate records. A case-cohort design was used. Five controls per term LBW case were randomly selected (without covariate matching or stratification) from among term births. The resulting datasets were analyzed by logistic regression with a random effect by hospital, using generalized additive mixed models adjusted for race/ethnicity, education, maternal age and household income. ResultsIn total 72,632 singleton term LBW cases were included. Term LBW was positively and significantly associated with interpolated measurements of ozone but not total fine PM or nitrogen dioxide. No significant association was observed between term LBW and primary PM from all sources grouped together. A positive significant association was observed for secondary organic aerosols. Exposure to elemental carbon (EC), nitrates and ammonium were also positively and significantly associated with term LBW, but only for exposure during the third trimester of pregnancy. Significant positive associations were observed between term LBW risk and primary PM emitted by on-road gasoline and diesel or by commercial meat cooking sources. Primary PM from wood burning was inversely associated with term LBW. Significant positive associations were also observed between term LBW and ultrafine particle numbers modeled with the line-source roadway dispersion model, traffic density and proximity to roadways. DiscussionThis large study based on complementary exposure metrics suggests that not only primary pollution sources (traffic and commercial meat cooking) but also EC and secondary pollutants are risk factors for term LBW.

Evolución de las inequidades en mortalidad por causas externas entre los municipios de Antioquia (Colombia)

ObjectiveTo analyse the trend in inequalities in mortality due to external causes among municipalities in Antioquia, department of Colombia, from 2000 to 2010, and its association with socioeconomic conditions. External causes included violent deaths, such as homicides, suicides and traffic accidents, among others. MethodsEcological design of mortality trends, with the 125 municipalities of Antioquia as the unit of analysis. The age-adjusted smoothed standardized mortality ratio (SMR) was estimated for each of the municipalities by using an empirical Bayesian model. Differences in the SMR between the poorest and least poor municipalities were estimated by using a two-level hierarchical model (level-1: year, level-2: municipality). ResultsMortality due to external causes showed a downward trend in the department in the period under review, although the situation was not similar in all municipalities. The findings showed that the risk of death from external causes significantly increased in poor and underdeveloped municipalities. ConclusionsIntervention is required through policies that take into account local differences in mortality due to external causes.

The placebo effect in thyroid cancer: a meta-analysis.

The natural history of advanced thyroid malignancies is largely unknown. The outcome of patients included in the placebo arm of clinical trials could be reflective of their therapy-free evolution. To analyze the response rate, symptoms and adverse effects of locally advanced or metastatic differentiated (DTC) and medullary thyroid cancer (MTC) in patients treated with placebo in clinical trials. PubMed (MEDLINE) and SCOPUS databases were searched through September 2015 to identify high-quality randomized controlled clinical trials. We included studies that recruited patients with DTC or MTC with a placebo arm. We conducted a meta-analysis for each category of response rate, adherence to treatment, and adverse events. An empirical Bayesian random-effect model was used. We identified five clinical trials. DTC and MTC were independently analyzed. In the placebo arm, no complete response was observed; partial response occurred in 1.6% (0.6-3) and 6.4% (3.4-10.3) of DTC and MTC respectively; stable disease was described in 40.5% (34.6-46.9) and 53.9% (44.3-64.4) of DTC and MTC respectively. DTC reached a disease control rate of 42.3% (36.2-48.9) and MTC of 60.2 (50.1-71.4). Treatment discontinuation rate was 3.5% (1.9-5.5) in DTC and 5.7% (3.0-9.4) in MTC. Rate of dose reduction was 7.3% (4.8-10.5) in DTC and 6.2% (3.3-10.0) in MTC. This meta-analysis provides extensive data on the response rate and adverse effects of locally advanced or metastatic DTC and MTC in patients treated with placebo. These results may be used for comparisons with results from clinical trials without a placebo arm.

An Integrated Approach to Empirical Bayesian Whole Genome Prediction Modeling

Computational efficiency is an increasing concern for whole genome prediction (WGP) based on denser genetic marker panels such that algorithms other than Markov Chain Monte Carlo (MCMC) warrant greater consideration, particularly for hierarchical models that flexibly confer either heavy-tailed (e.g., BayesA) or stochastic search and variable selection (SSVS) instead of Gaussian specifications on marker effect distributions. The expectation maximization (EM) algorithm is one attractive alternative; however, recently proposed hierarchical model implementations of EM have not addressed formal estimation of underlying hyperparameters even though their specifications are known to impact WGP accuracy. Furthermore, EM can be sensitive to starting values. We develop and explore the properties of an empirical Bayes strategy by conditioning EM implementations of BayesA or SSVS WGP models on marginal modal estimation of variance components and other key hyperparameters. These empirical Bayes implementations are compared against their MCMC counterparts for estimation of hyperparameters and WGP accuracy, both within the context of a simulation study and application to a loblolly pine dataset. In all cases, starting values were deemed to be important for EM-based estimates. Starting values based on MCMC posterior means were preferable, whereas those based on setting all marker effects equal to zero generally led to inferior performance. Nevertheless, a recently proposed regularization procedure was useful in alleviating the impact of starting values in the EM implementation of the SSVS model, as was modifying the expectation step in the BayesA model to be based on relative variances rather than on relative precisions.