Johannes M Freudenberg

Experimental comparison and cross-validation of the Affymetrix and Illumina gene expression analysis platforms

The growth in popularity of RNA expression microarrays has been accompanied by concerns about the reliability of the data especially when comparing between different platforms. Here, we present an evaluation of the reproducibility of microarray results using two platforms, Affymetrix GeneChips and Illumina BeadArrays. The study design is based on a dilution series of two human tissues (blood and placenta), tested in duplicate on each platform. The results of a comparison between the platforms indicate very high agreement, particularly for genes which are predicted to be differentially expressed between the two tissues. Agreement was strongly correlated with the level of expression of a gene. Concordance was also improved when probes on the two platforms could be identified as being likely to target the same set of transcripts of a given gene. These results shed light on the causes or failures of agreement across microarray platforms. The set of probes we found to be most highly reproducible can be used by others to help increase confidence in analyses of other data sets using these platforms.

Transcriptional recapitulation and subversion of embryonic colon development by mouse colon tumor models and human colon cancer

BackgroundThe expression of carcino-embryonic antigen by colorectal cancer is an example of oncogenic activation of embryonic gene expression. Hypothesizing that oncogenesis-recapitulating-ontogenesis may represent a broad programmatic commitment, we compared gene expression patterns of human colorectal cancers (CRCs) and mouse colon tumor models to those of mouse colon development embryonic days 13.5-18.5.ResultsWe report here that 39 colon tumors from four independent mouse models and 100 human CRCs encompassing all clinical stages shared a striking recapitulation of embryonic colon gene expression. Compared to normal adult colon, all mouse and human tumors over-expressed a large cluster of genes highly enriched for functional association to the control of cell cycle progression, proliferation, and migration, including those encoding MYC, AKT2, PLK1 and SPARC. Mouse tumors positive for nuclear β-catenin shifted the shared embryonic pattern to that of early development. Human and mouse tumors differed from normal embryonic colon by their loss of expression modules enriched for tumor suppressors (EDNRB, HSPE, KIT and LSP1). Human CRC adenocarcinomas lost an additional suppressor module (IGFBP4, MAP4K1, PDGFRA, STAB1 and WNT4). Many human tumor samples also gained expression of a coordinately regulated module associated with advanced malignancy (ABCC1, FOXO3A, LIF, PIK3R1, PRNP, TNC, TIMP3 and VEGF).ConclusionCross-species, developmental, and multi-model gene expression patterning comparisons provide an integrated and versatile framework for definition of transcriptional programs associated with oncogenesis. This approach also provides a general method for identifying pattern-specific biomarkers and therapeutic targets. This delineation and categorization of developmental and non-developmental activator and suppressor gene modules can thus facilitate the formulation of sophisticated hypotheses to evaluate potential synergistic effects of targeting within- and between-modules for next-generation combinatorial therapeutics and improved mouse models.

A semi-parametric Bayesian model for unsupervised differential co-expression analysis

BackgroundDifferential co-expression analysis is an emerging strategy for characterizing disease related dysregulation of gene expression regulatory networks. Given pre-defined sets of biological samples, such analysis aims at identifying genes that are co-expressed in one, but not in the other set of samples.ResultsWe developed a novel probabilistic framework for jointly uncovering contexts (i.e. groups of samples) with specific co-expression patterns, and groups of genes with different co-expression patterns across such contexts. In contrast to current clustering and bi-clustering procedures, the implicit similarity measure in this model used for grouping biological samples is based on the clustering structure of genes within each sample and not on traditional measures of gene expression level similarities. Within this framework, biological samples with widely discordant expression patterns can be placed in the same context as long as the co-clustering structure of genes is concordant within these samples. To the best of our knowledge, this is the first method to date for unsupervised differential co-expression analysis in this generality. When applied to the problem of identifying molecular subtypes of breast cancer, our method identified reproducible patterns of differential co-expression across several independent expression datasets. Sample groupings induced by these patterns were highly informative of the disease outcome. Expression patterns of differentially co-expressed genes provided new insights into the complex nature of the ERα regulatory network.ConclusionsWe demonstrated that the use of the co-clustering structure as the similarity measure in the unsupervised analysis of sample gene expression profiles provides valuable information about expression regulatory networks.

Influence of fatty acid diets on gene expression in rat mammary epithelial cells

BACKGROUND This study examines the impact of dietary fatty acids on regulation of gene expression in mammary epithelial cells before and during puberty. METHODS Diets primarily consisted of n-9 monounsaturated fatty acids (olive oil), n-6 polyunsaturated fatty acids (safflower), saturated acids (butter), and the reference AIN-93G diet (soy oil). The dietary regimen mimics the repetitive nature of fatty acid exposure in Western diets. Diet-induced changes in gene expression were examined in laser capture microdissected mammary ductal epithelial cells at day of weaning and end of puberty. PCNA immunohistochemistry analysis compared proliferation rates between diets. RESULTS Genes differentially expressed between each test diets and the reference diet were significantly enriched by cell cycle genes. Some of these genes were involved in activation of the cell cycle pathway or the G2/M check point pathway. Although there were some differences in the level of differential expression, all diets showed qualitatively the same pattern of differential expression compared to the reference diet. Cluster analysis identified an expanded set of cell cycle as well as immunity and sterol metabolism related clusters of differentially expressed genes. CONCLUSION Fatty acid-enriched diets significantly upregulated proliferation above normal physiological levels during puberty. Higher cellular proliferation during puberty caused by enriched fatty acid diets poses a potential increase risk of mammary cancer in later life. The human homologs of 27 of 62 cell cycle rat genes are included in a human breast cancer cluster of 45 cell cycle genes, further emphasizing the importance of our findings in the rat model.

Genome-Wide Signatures of Transcription Factor Activity: Connecting Transcription Factors, Disease, and Small Molecules

Identifying transcription factors (TF) involved in producing a genome-wide transcriptional profile is an essential step in building mechanistic model that can explain observed gene expression data. We developed a statistical framework for constructing genome-wide signatures of TF activity, and for using such signatures in the analysis of gene expression data produced by complex transcriptional regulatory programs. Our framework integrates ChIP-seq data and appropriately matched gene expression profiles to identify True REGulatory (TREG) TF-gene interactions. It provides genome-wide quantification of the likelihood of regulatory TF-gene interaction that can be used to either identify regulated genes, or as genome-wide signature of TF activity. To effectively use ChIP-seq data, we introduce a novel statistical model that integrates information from all binding “peaks” within 2 Mb window around a genes transcription start site (TSS), and provides gene-level binding scores and probabilities of regulatory interaction. In the second step we integrate these binding scores and regulatory probabilities with gene expression data to assess the likelihood of True REGulatory (TREG) TF-gene interactions. We demonstrate the advantages of TREG framework in identifying genes regulated by two TFs with widely different distribution of functional binding events (ERα and E2f1). We also show that TREG signatures of TF activity vastly improve our ability to detect involvement of ERα in producing complex diseases-related transcriptional profiles. Through a large study of disease-related transcriptional signatures and transcriptional signatures of drug activity, we demonstrate that increase in statistical power associated with the use of TREG signatures makes the crucial difference in identifying key targets for treatment, and drugs to use for treatment. All methods are implemented in an open-source R package treg. The package also contains all data used in the analysis including 494 TREG binding profiles based on ENCODE ChIP-seq data. The treg package can be downloaded at http://GenomicsPortals.org.

WebGimm: An integrated web-based platform for cluster analysis, functional analysis, and interactive visualization of results

Cluster analysis methods have been extensively researched, but the adoption of new methods is often hindered by technical barriers in their implementation and use. WebGimm is a free cluster analysis web-service, and an open source general purpose clustering web-server infrastructure designed to facilitate easy deployment of integrated cluster analysis servers based on clustering and functional annotation algorithms implemented in R. Integrated functional analyses and interactive browsing of both, clustering structure and functional annotations provides a complete analytical environment for cluster analysis and interpretation of results. The Java Web Start client-based interface is modeled after the familiar cluster/treeview packages making its use intuitive to a wide array of biomedical researchers. For biomedical researchers, WebGimm provides an avenue to access state of the art clustering procedures. For Bioinformatics methods developers, WebGimm offers a convenient avenue to deploy their newly developed clustering methods. WebGimm server, software and manuals can be freely accessed at http://ClusterAnalysis.org/.

Gene expression profiling and machine learning to understand and predict primary graft dysfunction

Lung transplantation is the treatment of choice for end-stage pulmonary diseases. A limited donor supply has resulted in 4000 patients on the waiting list. Currently, 10-20% of donor organs are deemed suitable under the selection criteria, of which 15-25% fails due to primary graft dysfunction (PGD). In this study, we attempt to further our understanding of PGD by observing the changes in gene expression across donor lungs that developed PGD versus those that did not. Our second goal is to use a machine learning tool - support vector machine (SVM), to distinguish unsuitable donor lungs from suitable donor lungs, based on the gene expression data. Classification results for distinguishing suitable and unsuitable lungs for transplantation using a SVM were promising. This is the first such attempt to use human lungs used for transplantation and combine the identification of a molecular signature for PGD, with machine learning methods for donor lung prediction.

A comprehensive analysis of the effect of microarray data

Background: Microarray data preprocessing, such as differentially expressed (DE) genes selection, is performed prior to higher level statistical analysis in order to account for technical variability. Preprocessing for the Affymetrix GeneChip includes background correction, normalisation and summarisation. Numerous preprocessing methods have been proposed with little consensus as to which is the most suitable. Furthermore, due to poor concordance among results from cross-platform analyses, protocols are being developed to enable cross-platform reproducibility. However, the effect of data analysis on a single platform is still unknown. The objective of our study is two-fold: first to determine whether there is consistency in the results obtained from a single platform; and second to investigate the effect of preprocessing on DE genes selection, analysed on four datasets. Results: Results indicate that microarray analysis is subjective. The lists of DE genes are variable and dependent on the preprocessing method used. Furthermore, the characteristics of the dataset, and the type of DE genes identification method used, greatly affect the outcome. Despite using a single platform, there is a lot of variability in the results. Conclusions: This is the first comprehensive analysis using multiple datasets generated from a single platform and involving many DE genes selection methods to assess the effect of data preprocessing on downstream analysis. Results indicate that preprocessing methods affect downstream analysis. Results are also affected by the kind of data and statistical analysis tools used. Our study reveals that there are inconsistencies in results obtained from a single platform. These issues have been overlooked in past reports. Type of Report: Other Department of Computer Science & Engineering Washington University in St. Louis Campus Box 1045 St. Louis, MO 63130 ph: (314) 935-6160 A comprehensive analysis of the effect of microarray data preprocessing methods on differentially expressed transcript