Volume 12, Issue 3

In genetic association studies (GAS) as well as in genome-wide association studies (GWAS), the mode of inheritance (dominant, additive and recessive) is usually not known a priori. Assuming an incorrect mode of inheritance may lead to substantial loss of power, whereas on the other hand, testing all possible models may result in an increased type I error rate. The situation is even more complicated in the meta-analysis of GAS or GWAS, in which individual studies are synthesized to derive an overall estimate. Meta-analysis increases the power to detect weak genotype effects, but heterogeneity and incompatibility between the included studies complicate things further. In this review, we present a comprehensive summary of the statistical methods used for robust analysis and genetic model selection in GAS and GWAS. We then discuss the application of such methods in the context of meta-analysis. We describe the theoretical properties of the various methods and the foundations on which they are based. We also present the available software implementations of the described methods. Finally, since only few of the available robust methods have been applied in the meta-analysis setting, we present some simple extensions that allow robust meta-analysis of GAS and GWAS. Possible extensions and proposals for future work are also discussed.

Novel uses of automated flow cytometry technology for measuring levels of protein markers on thousands to millions of cells are promoting increasing need for relevant, customized Bayesian mixture modelling approaches in many areas of biomedical research and application. In studies of immune profiling in many biological areas, traditional flow cytometry measures relative levels of abundance of marker proteins using fluorescently labeled tags that identify specific markers by a single-color. One specific and important recent development in this area is the use of combinatorial marker assays in which each marker is targeted with a probe that is labeled with two or more fluorescent tags. The use of several colors enables the identification of, in principle, combinatorially increasingly numbers of subtypes of cells, each identified by a subset of colors. This represents a major advance in the ability to characterize variation in immune responses involving larger numbers of functionally differentiated cell subtypes. We describe novel classes of Markov chain Monte Carlo methods for model fitting that exploit distributed GPU (graphics processing unit) implementation. We discuss issues of cellular subtype identification in this novel, general model framework, and provide a detailed example using simulated data. We then describe application to a data set from an experimental study of antigen-specific T-cell subtyping using combinatorially encoded assays in human blood samples. Summary comments discuss broader questions in applications in immunology, and aspects of statistical computation.

Classical approaches for analyzing reverse transcription quantitative polymerase chain reaction (RT-qPCR) data commonly require normalization before assessing differential expression (DE). Normalization often has a substantial effect on the interpretation and validity of the subsequent analysis steps, but at the same time it causes a reduction in variance and introduces dependence among the normalized outcomes. These effects can be substantial, however, they are typically ignored. Most normalization techniques and methods for DE focus on mean expression and are sensitive to outliers. Moreover, in cancer studies, for example, oncogenes are often only expressed in a subsample of the populations during sampling. This primarily affects the skewness and the tails of the distribution and the mean is therefore not necessarily the best effect size measure within these experimental setups. In our contribution, we propose an extension of the Wilcoxon-Mann-Whitney test which incorporates a robust normalization, and the uncertainty associated with normalization is propagated into the final statistical summaries for DE. Our method relies on semiparametric regression models that focus on the probability P { Y ≤ Y′ }, where Y and Y′ denote independent responses for different subject groups. This effect size is robust to outliers, while remaining informative and intuitive when DE affects the shape of the distribution instead of only the mean. We also extend our approach for assessing DE for multiple features simultaneously. Simulation studies show that the test has a good performance, and that it is very competitive with standard methods for this platform. The method is illustrated on two neuroblastoma studies.

High-throughput expression profiling allows simultaneous measure of tens of thousands of genes at once. These data have motivated the development of reliable biomarkers for disease subtypes identification and diagnosis. Many methods have been developed in the literature for analyzing these data, such as diagonal discriminant analysis, support vector machines, and k-nearest neighbor methods. The diagonal discriminant methods have been shown to perform well for high-dimensional data with small sample sizes. Despite its popularity, the independence assumption is unlikely to be true in practice. Recently, a gene module based linear discriminant analysis strategy has been proposed by utilizing the correlation among genes in discriminant analysis. However, the approach can be underpowered when the samples of the two classes are unbalanced. In this paper, we propose to correct the biases in the discriminant scores of block-diagonal discriminant analysis. In simulation studies, our proposed method outperforms other approaches in various settings. We also illustrate our proposed discriminant analysis method for analyzing microarray data studies.

The explosion of data in evolutionary bioinformatics has led to sometimes ad hoc, incomplete and even inaccurate data analyses. Taking dS data, namely, data on synonymous substitutions per synonymous sites, we go through a statistical analysis for modeling the time since duplications of genes. We explore the shortcomings of previous analyses, especially with a view towards their effect on inference for the gene duplication process. We present a statistical analysis which respects the assumptions of the models and the integrity of the data, and emphasize that exploratory data analysis, formulation of a data model, its estimation and finally, assessment of the model are important steps in a complete data analysis. Furthermore, for dS data, we develop Bayesian discrete-continuous mixture models and present analyses using two genomes.

Different statistical models have been proposed for maximizing prediction accuracy in genome-based prediction of breeding values in plant and animal breeding. However, little is known about the sensitivity of these models with respect to prior and hyperparameter specification, because comparisons of prediction performance are mainly based on a single set of hyperparameters. In this study, we focused on Bayesian prediction methods using a standard linear regression model with marker covariates coding additive effects at a large number of marker loci. By comparing different hyperparameter settings, we investigated the sensitivity of four methods frequently used in genome-based prediction (Bayesian Ridge, Bayesian Lasso, BayesA and BayesB) to specification of the prior distribution of marker effects. We used datasets simulated according to a typical maize breeding program differing in the number of markers and the number of simulated quantitative trait loci affecting the trait. Furthermore, we used an experimental maize dataset, comprising 698 doubled haploid lines, each genotyped with 56110 single nucleotide polymorphism markers and phenotyped as testcrosses for the two quantitative traits grain dry matter yield and grain dry matter content. The predictive ability of the different models was assessed by five-fold cross-validation. The extent of Bayesian learning was quantified by calculation of the Hellinger distance between the prior and posterior densities of marker effects. Our results indicate that similar predictive abilities can be achieved with all methods, but with BayesA and BayesB hyperparameter settings had a stronger effect on prediction performance than with the other two methods. Prediction performance of BayesA and BayesB suffered substantially from a non-optimal choice of hyperparameters.

In genomic analysis, there is growing interest in network structures that represent biochemistry interactions. Graph structured or constrained inference takes advantage of a known relational structure among variables to introduce smoothness and reduce complexity in modeling, especially for high-dimensional genomic data. There has been a lot of interest in its application in model regularization and selection. However, prior knowledge on the graphical structure among the variables can be limited and partial. Empirical data may suggest variations and modifications to such a graph, which could lead to new and interesting biological findings. In this paper, we propose a Bayesian random graph-constrained model, rGrace , an extension from the Grace model, to combine a priori network information with empirical evidence, for applications such as pathway analysis. Using both simulations and real data examples, we show that the new method, while leading to improved predictive performance, can identify discrepancy between data and a prior known graph structure and suggest modifications and updates.