Abstract Motivation: Selecting a small number of signature genes for accurate classification of samples is essential for the development of diagnostic tests. However, many genes are highly correlated in gene expression data, and hence, many possible sets of genes are potential classifiers. Because treatment outcomes are poor in advanced chronic myeloid leukemia (CML), we hypothesized that expression of classifiers of advanced phase CML when detected in early CML [chronic phase (CP) CML], correlates with subsequent poorer therapeutic outcome. Results: We developed a method that integrates gene expression data with expert knowledge and predicted functional relationships using iterative Bayesian model averaging. Applying our integrated method to CML, we identified small sets of signature genes that are highly predictive of disease phases and that are more robust and stable than using expression data alone. The accuracy of our algorithm was evaluated using cross-validation on the gene expression data. We then tested the hypothesis that gene sets associated with advanced phase CML would predict relapse after allogeneic transplantation in 176 independent CP CML cases. Our gene signatures of advanced phase CML are predictive of relapse even after adjustment for known risk factors associated with transplant outcomes. Availability: The source codes and data sets used are available from the web site http://expression.washington.edu/publications/kayee/integratedBMA. Contact: [email protected] Supplementary information: Supplementary data are available at Bioinformatics online.

BACKGROUND: Microarray technology is increasingly used to identify potential biomarkers for cancer prognostics and diagnostics. Previously, we have developed the iterative Bayesian Model Averaging (BMA) algorithm for use in classification. Here, we extend the iterative BMA algorithm for application to survival analysis on high-dimensional microarray data. The main goal in applying survival analysis to microarray data is to determine a highly predictive model of patients' time to event (such as death, relapse, or metastasis) using a small number of selected genes. Our multivariate procedure combines the effectiveness of multiple contending models by calculating the weighted average of their posterior probability distributions. Our results demonstrate that our iterative BMA algorithm for survival analysis achieves high prediction accuracy while consistently selecting a small and cost-effective number of predictor genes. RESULTS: We applied the iterative BMA algorithm to two cancer datasets: breast cancer and diffuse large B-cell lymphoma (DLBCL) data. On the breast cancer data, the algorithm selected a total of 15 predictor genes across 84 contending models from the training data. The maximum likelihood estimates of the selected genes and the posterior probabilities of the selected models from the training data were used to divide patients in the test (or validation) dataset into high- and low-risk categories. Using the genes and models determined from the training data, we assigned patients from the test data into highly distinct risk groups (as indicated by a p-value of 7.26e-05 from the log-rank test). Moreover, we achieved comparable results using only the 5 top selected genes with 100% posterior probabilities. On the DLBCL data, our iterative BMA procedure selected a total of 25 genes across 3 contending models from the training data. Once again, we assigned the patients in the validation set to significantly distinct risk groups (p-value = 0.00139). CONCLUSION: The strength of the iterative BMA algorithm for survival analysis lies in its ability to account for model uncertainty. The results from this study demonstrate that our procedure selects a small number of genes while eclipsing other methods in predictive performance, making it a highly accurate and cost-effective prognostic tool in the clinical setting.

MOTIVATION: Inner holes, artifacts and blank spots are common in microarray images, but current image analysis methods do not pay them enough attention. We propose a new robust model-based method for processing microarray images so as to estimate foreground and background intensities. The method starts with a very simple but effective automatic gridding method, and then proceeds in two steps. The first step applies model-based clustering to the distribution of pixel intensities, using the Bayesian Information Criterion (BIC) to choose the number of groups up to a maximum of three. The second step is spatial, finding the large spatially connected components in each cluster of pixels. The method thus combines the strengths of the histogram-based and spatial approaches. It deals effectively with inner holes in spots and with artifacts. It also provides a formal inferential basis for deciding when the spot is blank, namely when the BIC favors one group over two or three. RESULTS: We apply our methods for gridding and segmentation to cDNA microarray images from an HIV infection experiment. In these experiments, our method had better stability across replicates than a fixed-circle segmentation method or the seeded region growing method in the SPOT software, without introducing noticeable bias when estimating the intensities of differentially expressed genes. AVAILABILITY: spotSegmentation, an R language package implementing both the gridding and segmentation methods is available through the Bioconductor project (http://www.bioconductor.org). The segmentation method requires the contributed R package MCLUST for model-based clustering (http://cran.us.r-project.org). CONTACT: [email protected].

MOTIVATION: Selecting a small number of relevant genes for accurate classification of samples is essential for the development of diagnostic tests. We present the Bayesian model averaging (BMA) method for gene selection and classification of microarray data. Typical gene selection and classification procedures ignore model uncertainty and use a single set of relevant genes (model) to predict the class. BMA accounts for the uncertainty about the best set to choose by averaging over multiple models (sets of potentially overlapping relevant genes). RESULTS: We have shown that BMA selects smaller numbers of relevant genes (compared with other methods) and achieves a high prediction accuracy on three microarray datasets. Our BMA algorithm is applicable to microarray datasets with any number of classes, and outputs posterior probabilities for the selected genes and models. Our selected models typically consist of only a few genes. The combination of high accuracy, small numbers of genes and posterior probabilities for the predictions should make BMA a powerful tool for developing diagnostics from expression data. AVAILABILITY: The source codes and datasets used are available from our Supplementary website.

BACKGROUND: One of the primary tasks in analysing gene expression data is finding genes that are differentially expressed in different samples. Multiple testing issues due to the thousands of tests run make some of the more popular methods for doing this problematic. RESULTS: We propose a simple method, Normal Uniform Differential Gene Expression (NUDGE) detection for finding differentially expressed genes in cDNA microarrays. The method uses a simple univariate normal-uniform mixture model, in combination with new normalization methods for spread as well as mean that extend the lowess normalization of Dudoit, Yang, Callow and Speed (2002) 1. It takes account of multiple testing, and gives probabilities of differential expression as part of its output. It can be applied to either single-slide or replicated experiments, and it is very fast. Three datasets are analyzed using NUDGE, and the results are compared to those given by other popular methods: unadjusted and Bonferroni-adjusted t tests, Significance Analysis of Microarrays (SAM), and Empirical Bayes for microarrays (EBarrays) with both Gamma-Gamma and Lognormal-Normal models. CONCLUSION: The method gives a high probability of differential expression to genes known/suspected a priori to be differentially expressed and a low probability to the others. In terms of known false positives and false negatives, the method outperforms all multiple-replicate methods except for the Gamma-Gamma EBarrays method to which it offers comparable results with the added advantages of greater simplicity, speed, fewer assumptions and applicability to the single replicate case. An R package called nudge to implement the methods in this paper will be made available soon at http://www.bioconductor.org.

We propose a method for choosing the number of colors or true gray levels in an image; this allows fully automatic segmentation of images. Our underlying probability model is a hidden Markov random field. Each number of colors considered is viewed as corresponding to a statistical model for the image, and the resulting models are compared via approximate Bayes factors. The Bayes factors are approximated using BIC (Bayesian Information Criterion), where the required maximized likelihood is approximated by the Qian-Titterington (1991) pseudolikelihood. We call the resulting criterion PLIC (Pseudolikelihood Information Criterion). We also discuss a simpler approximation, MMIC (Marginal Mixture Information Criterion), which is based only on the marginal distribution of pixel values. This turns out to be useful for initialization and it also has moderately good performance by itself when the amount of spatial dependence in an image is low. We apply PLIC and MMIC to a medical image segmentation problem.

MOTIVATION: Clustering is a useful exploratory technique for the analysis of gene expression data. Many different heuristic clustering algorithms have been proposed in this context. Clustering algorithms based on probability models offer a principled alternative to heuristic algorithms. In particular, model-based clustering assumes that the data is generated by a finite mixture of underlying probability distributions such as multivariate normal distributions. The issues of selecting a 'good' clustering method and determining the 'correct' number of clusters are reduced to model selection problems in the probability framework. Gaussian mixture models have been shown to be a powerful tool for clustering in many applications. RESULTS: We benchmarked the performance of model-based clustering on several synthetic and real gene expression data sets for which external evaluation criteria were available. The model-based approach has superior performance on our synthetic data sets, consistently selecting the correct model and the number of clusters. On real expression data, the model-based approach produced clusters of quality comparable to a leading heuristic clustering algorithm, but with the key advantage of suggesting the number of clusters and an appropriate model. We also explored the validity of the Gaussian mixture assumption on different transformations of real data. We also assessed the degree to which these real gene expression data sets fit multivariate Gaussian distributions both before and after subjecting them to commonly used data transformations. Suitably chosen transformations seem to result in reasonable fits. AVAILABILITY: MCLUST is available at http://www.stat.washington.edu/fraley/mclust. The software for the diagonal model is under development. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: http://www.cs.washington.edu/homes/kayee/model.

Clustering about principal curves combines parametric modeling of noise with nonparametric modeling of feature shape. This is useful for detecting curvilinear features in spatial point patterns, with or without background noise. Applications include the detection of curvilinear minefields from reconnaissance images, some of the points in which represent false detections, and the detection of seismic faults from earthquake catalogs. Our algorithm for principal curve clustering is in two steps: The first is hierarchical and agglomerative (HPCC) and the second consists of iterative relocation based on the classification EM algorithm (CEM-PCC). HPCC is used to combine potential feature clusters, while CEM-PCC refines the results and deals with background noise. It is important to have a good starting point for the algorithm: This can be found manually or automatically using, for example, nearest neighbor clutter removal or model-based clustering. We choose the number of features and the amount of smoothing simultaneously, using approximate Bayes factors.

We consider the problem of determining the structure of clustered data, without prior knowledge of the number of clusters or any other information about their composition. Data are represented by a mixture model in which each component corresponds to a different cluster. Models with varying geometric properties are obtained through Gaussian components with different parametrizations and cross-cluster constraints. Noise and outliers can be modelled by adding a Poisson process component. Partitions are determined by the expectation-maximization (EM) algorithm for maximum likelihood, with initial values from agglomerative hierarchical clustering. Models are compared using an approximation to the Bayes factor based on the Bayesian information criterion (BIC); unlike significance tests, this allows comparison of more than two models at the same time, and removes the restriction that the models compared be nested. The problems of determining the number of clusters and the clustering method are solved simultaneously by choosing the best model. Moreover, the EM result provides a measure of uncertainty about the associated classification of each data point. Examples are given, showing that this approach can give performance that is much better than standard procedures, which often fail to identify groups that are either overlapping or of varying sizes and shapes.