Models for microarray gene expression data

Inflammatory and repair pathways induced in human bronchoalveolar lavage cells with ozone inhalation

Background

Inhalation of ambient levels of ozone causes airway inflammation and epithelial injury.

Methods

To examine the responses of airway cells to ozone-induced oxidative injury, 19 subjects (7 with asthma) were exposed to clean air (0ppb), medium (100ppb), and high (200ppb) ambient levels of ozone for 4h on three separate occasions in a climate-controlled chamber followed by bronchoscopy with bronchoalveolar lavage (BAL) 24h later. BAL cell mRNA expression was examined using Affymetrix GeneChip Microarray. The role of a differentially expressed gene (DEG) in epithelial injury was evaluated in an in vitro model of injury [16HBE14o- cell line scratch assay].

Results

Ozone exposure caused a dose-dependent up-regulation of several biologic pathways involved in inflammation and repair including chemokine and cytokine secretion, activity, and receptor binding; metalloproteinase and endopeptidase activity; adhesion, locomotion, and migration; and cell growth and tumorigenesis regulation. Asthmatic subjects had 1.7- to 3.8-fold higher expression of many DEGs suggestive of increased proinflammatory and matrix degradation and remodeling signals. The most highly up-regulated gene was osteopontin, the protein level of which in BAL fluid increased in a dose-dependent manner after ozone exposure. Asthmatic subjects had a disproportionate increase in non-polymerized osteopontin with increasing exposure to ozone. Treatment with polymeric, but not monomeric, osteopontin enhanced the migration of epithelial cells and wound closure in an α9β1 integrin-dependent manner.

Conclusions

Expression profiling of BAL cells after ozone exposure reveals potential regulatory genes and pathways activated by oxidative stress. One DEG, osteopontin, promotes epithelial wound healing in an in vitro model of injury.

Robustness considerations in selecting efficient two-color microarray designs

The main goal of microarray experiments is to select a small subset of genes that are differentially expressed among competing mRNA samples. For a given set of such mRNA samples, it is possible to consider a number of two-color cDNA microarray designs with a fixed number of arrays. Appropriate criteria can be used to select an efficient design from such a set of alternative experimental designs. In practice, however, microarray expression data often contain missing observations and the most efficient design (with complete observations) for a specific setup may not be efficient in the presence of missing observations. In this article, we propose two criteria to address the robustness of microarray designs against missing observations. We demonstrate the simultaneous use of efficiency and robustness criteria to select good microarray designs for both one-factor and multi-factor experiments.

Highly sensitive detection of melanoma based on serum proteomic profiling

PURPOSE

There is no available tumor marker that can detect primary melanoma. Proteomics analysis has been proposed as a novel tool that would lead to the discovery of potential new tumor markers.

METHODS

We developed a serum proteomic fingerprinting approach coupled with a classification method to determine whether proteomic profiling could discriminate between melanoma and healthy volunteers. A total of 108 serum samples from 30 early-stage [American Joint Committee on Cancer (AJCC) stage I or II] and 30 advanced-stage (AJCC stage III or IV) melanoma patients and 48 healthy volunteers were analyzed by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS) utilizing protein chip technology and artificial neural networks.

RESULTS

In a first step, a multiprotein classifier was built using a training set of 30 pathologically confirmed melanoma and 24 healthy volunteer serum samples, resulting in good classification accuracy for correct diagnosis and stage classification assignment. Subsequently, our multiprotein classifier was tested in an independent validation set of 30 melanoma and 24 non-cancer serum samples patients, maintained in a good diagnostic accuracy of 98.1% (sensitivity 96.7%, specificity 100%), and 100% stage I/II classification assignment.

CONCLUSIONS

Although results remain to be confirmed in larger collective patient cohorts, we could demonstrate the usefulness of proteomic profiling as a sensitive and specific assay to detect melanoma, including non-metastatic melanoma, from the serum.

Bayesian classification and non-Bayesian label estimation via EM algorithm to identify differentially expressed genes: a comparative study

Gene classification problem is studied considering the ratio of gene expression levels, X, in two-channel microarrays and a non-observed categorical variable indicating how differentially expressed the gene is: non differentially expressed, down-regulated or up-regulated. Supposing X from a mixture of Gamma distributions, two methods are proposed and results are compared. The first method is based on an hierarchical Bayesian model. The conditional predictive probability of a gene to belong to each group is calculated and the gene is assigned to the group for which this conditional probability is higher. The second method uses EM algorithm to estimate the most likely group label for each gene, that is, to assign the gene to the group which contains it with the higher estimated probability.

A marginal mixture model for selecting differentially expressed genes across two types of tissue samples

Bayesian hierarchical models that characterize the distributions of (transformed) gene profiles have been proven very useful and flexible in selecting differentially expressed genes across different types of tissue samples (e.g. Lo and Gottardo, 2007). However, the marginal mean and variance of these models are assumed to be the same for different gene clusters and for different tissue types. Moreover, it is not easy to determine which of the many competing Bayesian hierarchical models provides the best fit for a specific microarray data set. To address these two issues, we propose a marginal mixture model that directly models the marginal distribution of transformed gene profiles. Specifically, we approximate the marginal distributions of transformed gene profiles via a mixture of three-component multivariate Normal distributions, each component of which has the same structures of marginal mean vector and covariance matrix as those for Bayesian hierarchical models, but the values can differ. Based on the proposed model, a method is derived to select genes differentially expressed across two types of tissue samples. The derived gene selection method performs well on a real microarray data set and consistently has the best performance (based on class agreement indices) compared with several other gene selection methods on simulated microarray data sets generated from three different mixture models.

A new class of mixture models for differential gene expression in DNA microarray data

One of the fundamental issues in analyzing microarray data is to determine which genes are expressed and which ones are not for a given group of subjects. In datasets where many genes are expressed and many are not expressed (i.e., underexpressed), a bimodal distribution for the gene expression levels often results, where one mode of the distribution represents the expressed genes and the other mode represents the underexpressed genes. To model this bimodality, we propose a new class of mixture models that utilize a random threshold value for accommodating bimodality in the gene expression distribution. Theoretical properties of the proposed model are carefully examined. We use this new model to examine the problem of differential gene expression between two groups of subjects, develop prior distributions, and derive a new criterion for determining which genes are differentially expressed between the two groups. Prior elicitation is carried out using empirical Bayes methodology in order to estimate the threshold value as well as elicit the hyperparameters for the two component mixture model. The new gene selection criterion is demonstrated via several simulations to have excellent false positive rate and false negative rate properties. A gastric cancer dataset is used to motivate and illustrate the proposed methodology.

Split-plot microarray experiments: issues of design, power and sample size

This article focuses on microarray experiments with two or more factors in which treatment combinations of the factors corresponding to the samples paired together onto arrays are not completely random. A main effect of one (or more) factor(s) is confounded with arrays (the experimental blocks). This is called a split-plot microarray experiment. We utilise an analysis of variance (ANOVA) model to assess differentially expressed genes for between-array and within-array comparisons that are generic under a split-plot microarray experiment. Instead of standard t- or F-test statistics that rely on mean square errors of the ANOVA model, we use a robust method, referred to as 'a pooled percentile estimator', to identify genes that are differentially expressed across different treatment conditions. We illustrate the design and analysis of split-plot microarray experiments based on a case application described by Jin et al. A brief discussion of power and sample size for split-plot microarray experiments is also presented.

Development of a "reverse capture" autoantibody microarray for studies of antigen-autoantibody profiling

Diagnosing cancers based on serum profiling is a particularly attractive concept. However, the technical challenges to analysis of the serum proteome arise from the dynamic range of protein amounts. Cancer sera contain antibodies that react with a unique group of autologous cellular antigens, which affords a dramatic amplification of signal in the form of antibodies relative to the amount of the corresponding antigens. The serum autoantibody repertoire from cancer patients might, therefore, be exploited for antigen-antibody profiling. To date, studies of antigen-antibody reactivity using microarrays have relied on recombinant proteins or synthetic peptides as arrayed features. However, recombinant proteins and/or synthetic peptides may fail to accurately detect autoantibody binding due to the lack of proper PTMs. Here we describe the development and use of a "reverse capture" autoantibody microarray. Our "reverse capture" autoantibody microarray is based on the dual-antibody sandwich immunoassay platform of ELISA, which allows the antigens to be immobilized in their native configuration. As "proof-of-principle", we demonstrate its use for antigen-autoantibody profiling with sera from patients with prostate cancer and benign prostate hyperplasia.

[11] Analysis of Variance of Microarray Data

Analysis of variance (ANOVA) is an approach used to identify differentially expressed genes in complex experimental designs. It is based on testing for the significance of the magnitude of effect of two or more treatments taking into account the variance within and between treatment classes. ANOVA is a highly flexible analytical approach that allows investigators to simultaneously assess the contributions of multiple factors to gene expression variation, including technical (dye, batch) effects and biological (sex, genotype, drug, time) ones, as well as interactions between factors. This chapter provides an overview of the theory of linear mixture modeling and the sequence of steps involved in fitting gene-specific models and discusses essential features of experimental design. Commercial and open-source software for performing ANOVA is widely available.

Nonparametric methods for microarray data based on exchangeability and borrowed power

This article proposes nonparametric inference procedures for analyzing microarray gene expression data that are reliable, robust, and simple to implement. They are conceptually transparent and require no special-purpose software. The analysis begins by normalizing gene expression data in a unique way. The resulting adjusted observations consist of gene-treatment interaction terms (representing differential expression) and error terms. The error terms are considered to be exchangeable, which is the only substantial assumption. Thus, under a family null hypothesis of no differential expression, the adjusted observations are exchangeable and all permutations of the observations are equally probable. The investigator may use the adjusted observations directly in a distribution-free test method or use their ranks in a rank-based method, where the ranking is taken over the whole data set. For the latter, the essential steps are as follows: (1) Calculate a Wilcoxon rank-sum difference or a corresponding Kruskal-Wallis rank statistic for each gene. (2) Randomly permute the observations and repeat the previous step. (3) Independently repeat the random permutation a suitable number of times. Under the exchangeability assumption, the permutation statistics are independent random draws from a null cumulative distribution function (c.d.f) approximated by the empirical c.d.f Reference to the empirical c.d.f tells if the test statistic for a gene is outlying and, hence, shows differential expression. This feature is judged by using an appropriate rejection region or computing a p-value for each test statistic, taking into account multiple testing. The distribution-free analog of the rank-based approach is also available and has parallel steps which are described in the article. The proposed nonparametric analysis tends to give good results with no additional refinement, although a few refinements are presented that may interest some investigators. The implementation is illustrated with a case application involving differential gene expression in wild-type and knockout mice of an E. coli lipopolysaccharide (LPS) endotoxin treatment, relative to a baseline untreated condition.

Analysis of a splice array experiment elucidates roles of chromatin elongation factor Spt4-5 in splicing

Splicing is an important process for regulation of gene expression in eukaryotes, and it has important functional links to other steps of gene expression. Two examples of these linkages include Ceg1, a component of the mRNA capping enzyme, and the chromatin elongation factors Spt4-5, both of which have recently been shown to play a role in the normal splicing of several genes in the yeast Saccharomyces cerevisiae. Using a genomic approach to characterize the roles of Spt4-5 in splicing, we used splicing-sensitive DNA microarrays to identify specific sets of genes that are mis-spliced in ceg1, spt4, and spt5 mutants. In the context of a complex, nested, experimental design featuring 22 dye-swap array hybridizations, comprising both biological and technical replicates, we applied five appropriate statistical models for assessing differential expression between wild-type and the mutants. To refine selection of differential expression genes, we then used a robust model-synthesizing approach, Differential Expression via Distance Synthesis, to integrate all five models. The resultant list of differentially expressed genes was then further analyzed with regard to select attributes: we found that highly transcribed genes with long introns were most sensitive to spt mutations. QPCR confirmation of differential expression was established for the limited number of genes evaluated. In this paper, we showcase splicing array technology, as well as powerful, yet general, statistical methodology for assessing differential expression, in the context of a real, complex experimental design. Our results suggest that the Spt4-Spt5 complex may help coordinate splicing with transcription under conditions that present kinetic challenges to spliceosome assembly or function.

Empirical bayes gene screening tool for time-course or dose-response microarray data

An efficient method to reduce the dimensionality of microarray gene expression data from thousands or tens of thousands of cDNA clones down to a subset of the most differentially expressed cDNA clones is essential in order to simplify the massive amount of data generated from microarray experiments. An extension to the methods of Efron et al. [Efron, B., Tibshirani, R., Storey, J., Tusher, V. (2001). Empirical Bayes analysis of a microarray experiment. J. Am. Statist. Assoc. 96:1151-1160] is applied to a differential time-course experiment to determine a subset of cDNAs that have the largest probability of being differentially expressed with respect to treatment conditions across a set of unequally spaced time points. The proposed extension, which is advocated to be a screening tool, allows for inference across a continuous variable in addition to incorporating a more complex experimental design and allowing for multiple design replications. With the current data the focus is on a time-course experiment; however, the proposed methods can easily be implemented on a dose-response experiment, or any other microarray experiment that contains a continuous variable of interest. The proposed empirical Bayes gene-screening tool is compared with the Efron et al. (2001) method in addition to an adjusted model-based t-value using a time-course data set where the toxicological effect of a specific mixture of chemicals is being studied.

Linear models for microarray data analysis: hidden similarities and differences

In the past several years many linear models have been proposed for analyzing two-color microarray data. As presented in the literature, many of these models appear dramatically different. However, many of these models are reformulations of the same basic approach to analyzing microarray data. This paper demonstrates the equivalence of some of these models. Attention is directed at choices in microarray data analysis that have a larger impact on the results than the choice of linear model.

Empirical Bayes estimation of gene-specific effects in micro-array research

Micro-array technology allows investigators the opportunity to measure expression levels of thousands of genes simultaneously. However, investigators are also faced with the challenge of simultaneous estimation of gene expression differences for thousands of genes with very small sample sizes. Traditional estimators of differences between treatment means (ordinary least squares estimators or OLS) are not the best estimators if interest is in estimation of gene expression differences for an ensemble of genes. In the case that gene expression differences are regarded as exchangeable samples from a common population, estimators are available that result in much smaller average mean-square error across the population of gene expression difference estimates. We have simulated the application of such an estimator, namely an empirical Bayes (EB) estimator of random effects in a hierarchical linear model (normal-normal). Simulation results revealed mean-square error as low as 0.05 times the mean-square error of OLS estimators (i.e., the difference between treatment means). We applied the analysis to an example dataset as a demonstration of the shrinkage of EB estimators and of the reduction in mean-square error, i.e., increase in precision, associated with EB estimators in this analysis. The method described here is available in software that is available at http://www.soph.uab.edu/ssg.asp?id=1087.

Genomic portraits of the nervous system in health and disease

As the human genome project moves toward its goal of sequencing the entire human genome, gene expression profiling by DNA microarray technology is being employed to rapidly screen genes for biological information. In this review, we will introduce DNA microarray technology, outline the basic experimental paradigms and data analysis methods, and then show with some examples how gene expression profiling can be applied to the study of the central nervous system in health and disease.

Effects of clofibric acid on mRNA expression profiles in primary cultures of rat, mouse and human hepatocytes

The mRNA expression profile in control and clofibric acid (CLO)-treated mouse, rat, and human hepatocytes was analyzed using species-specific oligonucleotide DNA microarrays (Affymetrix). A statistical empirical Bayes procedure was applied in order to select the significantly differentially expressed genes. Treatment with the peroxisome proliferator CLO induced up-regulation of genes involved in peroxisome proliferation and in cell proliferation as well as down-regulation of genes involved in apoptosis in hepatocytes of rodent but not of human origin. CLO treatment induced up-regulation of microsomal cytochrome P450 4a genes in rodent hepatocytes and in two of six human hepatocyte cultures. In addition, genes encoding phenobarbital-inducible cytochrome P450s were also up-regulated by CLO in rodent and human hepatocyte cultures. Up-regulation of phenobarbital-inducible UDP-glucuronosyl-transferase genes by CLO was observed in both rat and human but not in mouse hepatocytes. CLO treatment induced up-regulation of L-fatty acid binding protein (L-FABP) gene in hepatocytes of both rodent and human origin. However, while genes of the cytosolic, microsomal, and mitochondrial pathways involved in fatty acid transport and metabolism were up-regulated by CLO in both rodent and human hepatocyte cultures, genes of the peroxisomal pathway of lipid metabolism were up-regulated in rodents only. An up-regulation of hepatocyte nuclear factor 1alpha (HNF1alpha) by CLO was observed only in human hepatocyte cultures, suggesting that this trans-activating factor may play a key role in the regulation of fatty acid metabolism in human liver as well as in the nonresponsiveness of human liver to CLO-induced regulation of cell proliferation and apoptosis.

Statistical tests for differential expression in cDNA microarray experiments

Extracting biological information from microarray data requires appropriate statistical methods. The simplest statistical method for detecting differential expression is the t test, which can be used to compare two conditions when there is replication of samples. With more than two conditions, analysis of variance (ANOVA) can be used, and the mixed ANOVA model is a general and powerful approach for microarray experiments with multiple factors and/or several sources of variation.

Power and sample size for DNA microarray studies

A microarray study aims at having a high probability of declaring genes to be differentially expressed if they are truly expressed, while keeping the probability of making false declarations of expression acceptably low. Thus, in formal terms, well-designed microarray studies will have high power while controlling type I error risk. Achieving this objective is the purpose of this paper. Here, we discuss conceptual issues and present computational methods for statistical power and sample size in microarray studies, taking account of the multiple testing that is generic to these studies. The discussion encompasses choices of experimental design and replication for a study. Practical examples are used to demonstrate the methods. The examples show forcefully that replication of a microarray experiment can yield large increases in statistical power. The paper refers to cDNA arrays in the discussion and illustrations but the proposed methodology is equally applicable to expression data from oligonucleotide arrays.