Ka Yee Yeung

Principal component analysis for clustering gene expression data

MOTIVATION There is a great need to develop analytical methodology to analyze and to exploit the information contained in gene expression data. Because of the large number of genes and the complexity of biological networks, clustering is a useful exploratory technique for analysis of gene expression data. Other classical techniques, such as principal component analysis (PCA), have also been applied to analyze gene expression data. Using different data analysis techniques and different clustering algorithms to analyze the same data set can lead to very different conclusions. Our goal is to study the effectiveness of principal components (PCs) in capturing cluster structure. Specifically, using both real and synthetic gene expression data sets, we compared the quality of clusters obtained from the original data to the quality of clusters obtained after projecting onto subsets of the principal component axes. RESULTS Our empirical study showed that clustering with the PCs instead of the original variables does not necessarily improve, and often degrades, cluster quality. In particular, the first few PCs (which contain most of the variation in the data) do not necessarily capture most of the cluster structure. We also showed that clustering with PCs has different impact on different algorithms and different similarity metrics. Overall, we would not recommend PCA before clustering except in special circumstances.

Validating clustering for gene expression data

MOTIVATION Many clustering algorithms have been proposed for the analysis of gene expression data, but little guidance is available to help choose among them. We provide a systematic framework for assessing the results of clustering algorithms. Clustering algorithms attempt to partition the genes into groups exhibiting similar patterns of variation in expression level. Our methodology is to apply a clustering algorithm to the data from all but one experimental condition. The remaining condition is used to assess the predictive power of the resulting clusters-meaningful clusters should exhibit less variation in the remaining condition than clusters formed by chance. RESULTS We successfully applied our methodology to compare six clustering algorithms on four gene expression data sets. We found our quantitative measures of cluster quality to be positively correlated with external standards of cluster quality.

Bayesian model averaging: development of an improved multi-class, gene selection and classification tool for microarray data

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.

Clustering gene-expression data with repeated measurements

Clustering is a common methodology for the analysis of array data, and many research laboratories are generating array data with repeated measurements. We evaluated several clustering algorithms that incorporate repeated measurements, and show that algorithms that take advantage of repeated measurements yield more accurate and more stable clusters. In particular, we show that the infinite mixture model-based approach with a built-in error model produces superior results.

Multiclass classification of microarray data with repeated measurements: application to cancer

Prediction of the diagnostic category of a tissue sample from its gene-expression profile and selection of relevant genes for class prediction have important applications in cancer research. We have developed the uncorrelated shrunken centroid (USC) and error-weighted, uncorrelated shrunken centroid (EWUSC) algorithms that are applicable to microarray data with any number of classes. We show that removing highly correlated genes typically improves classification results using a small set of genes.

Donuts, scratches and blanks: robust model-based segmentation of microarray images

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 fraley@stat.washington.edu.

From co-expression to co-regulation: how many microarray experiments do we need?

BackgroundCluster analysis is often used to infer regulatory modules or biological function by associating unknown genes with other genes that have similar expression patterns and known regulatory elements or functions. However, clustering results may not have any biological relevance.ResultsWe applied various clustering algorithms to microarray datasets with different sizes, and we evaluated the clustering results by determining the fraction of gene pairs from the same clusters that share at least one known common transcription factor. We used both yeast transcription factor databases (SCPD, YPD) and chromatin immunoprecipitation (ChIP) data to evaluate our clustering results. We showed that the ability to identify co-regulated genes from clustering results is strongly dependent on the number of microarray experiments used in cluster analysis and the accuracy of these associations plateaus at between 50 and 100 experiments on yeast data. Moreover, the model-based clustering algorithm MCLUST consistently outperforms more traditional methods in accurately assigning co-regulated genes to the same clusters on standardized data.ConclusionsOur results are consistent with respect to independent evaluation criteria that strengthen our confidence in our results. However, when one compares ChIP data to YPD, the false-negative rate is approximately 80% using the recommended p-value of 0.001. In addition, we showed that even with large numbers of experiments, the false-positive rate may exceed the true-positive rate. In particular, even when all experiments are included, the best results produce clusters with only a 28% true-positive rate using known gene transcription factor interactions.

Construction of regulatory networks using expression time-series data of a genotyped population

The inference of regulatory and biochemical networks from large-scale genomics data is a basic problem in molecular biology. The goal is to generate testable hypotheses of gene-to-gene influences and subsequently to design bench experiments to confirm these network predictions. Coexpression of genes in large-scale gene-expression data implies coregulation and potential gene–gene interactions, but provide little information about the direction of influences. Here, we use both time-series data and genetics data to infer directionality of edges in regulatory networks: time-series data contain information about the chronological order of regulatory events and genetics data allow us to map DNA variations to variations at the RNA level. We generate microarray data measuring time-dependent gene-expression levels in 95 genotyped yeast segregants subjected to a drug perturbation. We develop a Bayesian model averaging regression algorithm that incorporates external information from diverse data types to infer regulatory networks from the time-series and genetics data. Our algorithm is capable of generating feedback loops. We show that our inferred network recovers existing and novel regulatory relationships. Following network construction, we generate independent microarray data on selected deletion mutants to prospectively test network predictions. We demonstrate the potential of our network to discover de novo transcription-factor binding sites. Applying our construction method to previously published data demonstrates that our method is competitive with leading network construction algorithms in the literature.

MeV+R: using MeV as a graphical user interface for Bioconductor applications in microarray analysis.

We present MeV+R, an integration of the JAVA MultiExperiment Viewer program with Bioconductor packages. This integration of MultiExperiment Viewer and R is easily extensible to other R packages and provides users with point and click access to traditionally command line driven tools written in R. We demonstrate the ability to use MultiExperiment Viewer as a graphical user interface for Bioconductor applications in microarray data analysis by incorporating three Bioconductor packages, RAMA, BRIDGE and iterativeBMA.

Fast Bayesian inference for gene regulatory networks using ScanBMA

BackgroundGenome-wide time-series data provide a rich set of information for discovering gene regulatory relationships. As genome-wide data for mammalian systems are being generated, it is critical to develop network inference methods that can handle tens of thousands of genes efficiently, provide a systematic framework for the integration of multiple data sources, and yield robust, accurate and compact gene-to-gene relationships.ResultsWe developed and applied ScanBMA, a Bayesian inference method that incorporates external information to improve the accuracy of the inferred network. In particular, we developed a new strategy to efficiently search the model space, applied data transformations to reduce the effect of spurious relationships, and adopted the g-prior to guide the search for candidate regulators. Our method is highly computationally efficient, thus addressing the scalability issue with network inference. The method is implemented as the ScanBMA function in the networkBMA Bioconductor software package.ConclusionsWe compared ScanBMA to other popular methods using time series yeast data as well as time-series simulated data from the DREAM competition. We found that ScanBMA produced more compact networks with a greater proportion of true positives than the competing methods. Specifically, ScanBMA generally produced more favorable areas under the Receiver-Operating Characteristic and Precision-Recall curves than other regression-based methods and mutual-information based methods. In addition, ScanBMA is competitive with other network inference methods in terms of running time.