Volume 13, Issue 4

Identifying population genetic structure (PGS) is crucial for breeding and conservation. Several clustering algorithms are available to identify the underlying PGS to be used with genetic data of maize genotypes. In this work, six methods to identify PGS from unlinked molecular marker data were compared using simulated and experimental data consisting of multilocus-biallelic genotypes. Datasets were delineated under different biological scenarios characterized by three levels of genetic divergence among populations (low, medium, and high F ST ) and two numbers of sub-populations ( K =3 and K =5). The relative performance of hierarchical and non-hierarchical clustering, as well as model-based clustering (STRUCTURE) and clustering from neural networks (SOM-RP-Q). We use the clustering error rate of genotypes into discrete sub-populations as comparison criterion. In scenarios with great level of divergence among genotype groups all methods performed well. With moderate level of genetic divergence ( F ST =0.2), the algorithms SOM-RP-Q and STRUCTURE performed better than hierarchical and non-hierarchical clustering. In all simulated scenarios with low genetic divergence and in the experimental SNP maize panel (largely unlinked), SOM-RP-Q achieved the lowest clustering error rate. The SOM algorithm used here is more effective than other evaluated methods for sparse unlinked genetic data.

There has been much recent interest in systems biology for investigating the structure of gene regulatory systems. Such networks are often formed of specific patterns, or network motifs, that are interesting from a biological point of view. Our aim in the present paper is to compare statistical methods specifically with regard to the question of how well they can detect such motifs. One popular approach is by network analysis with Gaussian graphical models (GGMs), which are statistical models associated with undirected graphs, where vertices of the graph represent genes and edges indicate regulatory interactions. Gene expression microarray data allow us to observe the amount of mRNA simultaneously for a large number of genes p under different experimental conditions n , where p is usually much larger than n prohibiting the use of standard methods. We therefore compare the performance of a number of procedures that have been specifically designed to address this large p -small n issue: G-Lasso estimation, Neighbourhood selection, Shrinkage estimation using empirical Bayes for model selection, and PC-algorithm. We found that all approaches performed poorly on the benchmark E. coli network. Hence we systematically studied their ability to detect specific network motifs, pairs, hubs and cascades, in extensive simulations. We conclude that all methods have difficulty detecting hubs, but the PC-algorithm is most promising.

DNA barcodes are short strands of 255–700 nucleotide bases taken from the cytochrome c oxidase subunit 1 (COI) region of the mitochondrial DNA. It has been proposed that these barcodes may be used as a method of differentiating between biological species. Current methods of species classification utilize distance measures that are heavily dependent on both evolutionary model assumptions as well as a clearly defined “gap” between intra- and interspecies variation. Such distance measures fail to measure classification uncertainty or to indicate how much of the barcode is necessary for classification. We propose a sequential naïve Bayes classifier for species classification to address these limitations. The proposed method is shown to provide accurate species-level classification on real and simulated data. The method proposed here quantifies the uncertainty of each classification and addresses how much of the barcode is necessary.

Pathway analysis has become a central approach to understanding the underlying biology of differentially expressed genes. As large amounts of microarray data have been accumulated in public repositories, flexible methodologies are needed to extend the analysis of simple case-control studies in order to place them in context with the vast quantities of available and highly heterogeneous data sets. To address this challenge, we have developed a two-level model, consisting of 1) a joint Bayesian factor model that integrates multiple microarray experiments and ties each factor to a predefined pathway and 2) a point mass mixture distribution that infers which factors are relevant/irrelevant to each dataset. Our method can identify pathways specific to a particular experimental trait which are concurrently induced/repressed under a variety of interventions. In this paper, we describe the model in depth and provide examples of its utility in simulations as well as real data from a study of radiation exposure. Our analysis of the radiation study leads to novel insights into the molecular basis of time- and dose- dependent response to ionizing radiation in mice peripheral blood. This broadly applicable model provides a starting point for generating specific and testable hypotheses in a pathway-centric manner.

The popular likelihood-based model selection criterion, Akaike’s Information Criterion (AIC), is a breakthrough mathematical result derived from information theory. AIC is an approximation to Kullback-Leibler (KL) divergence with the derivation relying on the assumption that the likelihood function has finite second derivatives. However, for phylogenetic estimation, given that tree space is discrete with respect to tree topology, the assumption of a continuous likelihood function with finite second derivatives is violated. In this paper, we investigate the relationship between the expected log likelihood of a candidate model, and the expected KL divergence in the context of phylogenetic tree estimation. We find that given the tree topology, AIC is an unbiased estimator of the expected KL divergence. However, when the tree topology is unknown, AIC tends to underestimate the expected KL divergence for phylogenetic models. Simulation results suggest that the degree of underestimation varies across phylogenetic models so that even for large sample sizes, the bias of AIC can result in selecting a wrong model. As the choice of phylogenetic models is essential for statistical phylogenetic inference, it is important to improve the accuracy of model selection criteria in the context of phylogenetics.

As the complexity and heterogeneity of cancer is being increasingly appreciated through genomic analyses, microarray-based cancer classification comprising multiple discriminatory molecular markers is an emerging trend. Such multiclass classification problems pose new methodological and computational challenges for developing novel and effective statistical approaches. In this paper, we introduce a new approach for classifying multiple disease states associated with cancer based on gene expression profiles. Our method focuses on detecting small sets of genes in which the relative comparison of their expression values leads to class discrimination. For an m -class problem, the classification rule typically depends on a small number of m -gene sets, which provide transparent decision boundaries and allow for potential biological interpretations. We first test our approach on seven common gene expression datasets and compare it with popular classification methods including support vector machines and random forests. We then consider an extremely large cohort of leukemia cancer patients to further assess its effectiveness. In both experiments, our method yields comparable or even better results to benchmark classifiers. In addition, we demonstrate that our approach can integrate pathway analysis of gene expression to provide accurate and biological meaningful classification.

Hierarchically-arranged multiple sequence alignment profiles are useful for modeling protein domains that have functionally diverged into evolutionarily-related subgroups. Currently such alignment hierarchies are largely constructed through manual curation, as for the NCBI Conserved Domain Database (CDD). Recently, however, I developed a Gibbs sampler that uses an approach termed statistical evolutionary dynamics analysis to accomplish this task automatically while, at the same time, identifying sequence determinants of protein function. Here I describe the statistical model and sampling strategies underlying this sampler. When implemented and applied to simulated protein sequences (which conform to the underlying statistical model precisely), these sampling strategies efficiently converge on the hierarchy used to generate the sequences. However, for real protein sequences the sampler finds alternative, nearly-optimal hierarchies for many domains, indicating a significant degree of ambiguity. I illustrate how both the nature of such ambiguities and the most robust (“consensus”) features of a hierarchy may be determined from an ensemble of independently generated hierarchies for the same domain. Such consensus hierarchies can provide reliably stable models of protein domain functional divergence.