Refine
Has Fulltext
- yes (2)
Is part of the Bibliography
- yes (2)
Document Type
- Journal article (1)
- Doctoral Thesis (1)
Language
- English (2) (remove)
Keywords
- classification (2) (remove)
Institute
- Theodor-Boveri-Institut für Biowissenschaften (2) (remove)
Recently, several classifiers that combine primary tumor data, like gene expression data, and secondary data sources, such as protein-protein interaction networks, have been proposed for predicting outcome in breast cancer. In these approaches, new composite features are typically constructed by aggregating the expression levels of several genes. The secondary data sources are employed to guide this aggregation. Although many studies claim that these approaches improve classification performance over single genes classifiers, the gain in performance is difficult to assess. This stems mainly from the fact that different breast cancer data sets and validation procedures are employed to assess the performance. Here we address these issues by employing a large cohort of six breast cancer data sets as benchmark set and by performing an unbiased evaluation of the classification accuracies of the different approaches. Contrary to previous claims, we find that composite feature classifiers do not outperform simple single genes classifiers. We investigate the effect of (1) the number of selected features; (2) the specific gene set from which features are selected; (3) the size of the training set and (4) the heterogeneity of the data set on the performance of composite feature and single genes classifiers. Strikingly, we find that randomization of secondary data sources, which destroys all biological information in these sources, does not result in a deterioration in performance of composite feature classifiers. Finally, we show that when a proper correction for gene set size is performed, the stability of single genes sets is similar to the stability of composite feature sets. Based on these results there is currently no reason to prefer prognostic classifiers based on composite features over single genes classifiers for predicting outcome in breast cancer.
In this thesis, the development of a phylogenetic DNA microarray, the analysis of several gene expression microarray datasets and new approaches for improved data analysis and interpretation are described. In the first publication, the development and analysis of a phylogenetic microarray is presented. I could show that species detection with phylogenetic DNA microarrays can be significantly improved when the microarray data is analyzed with a linear regression modeling approach. Standard methods have so far relied on pure signal intensities of the array spots and a simple cutoff criterion was applied to call a species present or absent. This procedure is not applicable to very closely related species with high sequence similarity because cross-hybridization of non-target DNA renders species detection impossible based on signal intensities alone. By modeling hybridization and cross-hybridization with linear regression, as I have presented in this thesis, even species with a sequence similarity of 97% in the marker gene can be detected and distinguished from related species. Another advantage of the modeling approach over existing methods is that the model also performs well on mixtures of different species. In principle, also quantitative predictions can be made. To make better use of the large amounts of microarray data stored in public databases, meta-analysis approaches need to be developed. In the second publication, an explorative meta-analysis exemplified on Arabidopsis thaliana gene expression datasets is presented. Integrating datasets studying effects such as the influence of plant hormones, pathogens and different mutations on gene expression levels, clusters of similarly treated datasets could be found. From the clusters of pathogen-treated and indole-3-acetic acid (IAA) treated datasets, representative genes were selected which pointed to functions which had been associated with pathogen attack or IAA effects previously. Additionally, hypotheses about the functions of so far uncharacterized genes could be set up. Thus, this kind of meta-analysis could be used to propose gene functions and their regulation under different conditions. In this work, also primary data analysis of Arabidopsis thaliana datasets is presented. In the third publication, an experiment which was conducted to find out if microwave irradiation has an effect on the gene expression of a plant cell culture is described. During the first steps, the data analysis was carried out blinded and exploratory analysis methods were applied to find out if the irradiation had an effect on gene expression of plant cells. Small but statistically significant changes in a few genes were found and could be experimentally confirmed. From the functions of the regulated genes and a meta-analysis with publicly available microarray data, it could be suspected that the plant cell culture somehow perceived the irradiation as energy, similar to perceiving light rays. The fourth publication describes the functional analysis of another Arabidopsis thaliana gene expression dataset. The gene expression data of the plant tumor dataset pointed to a switch from a mainly aerobic, auxotrophic to an anaerobic and heterotrophic metabolism in the plant tumor. Genes involved in photosynthesis were found to be repressed in tumors; genes of amino acid and lipid metabolism, cell wall and solute transporters were regulated in a way that sustains tumor growth and development. Furthermore, in the fifth publication, GEPAT (Genome Expression Pathway Analysis Tool), a tool for the analysis and integration of microarray data with other data types, is described. It consists of a web application and database which allows comfortable data upload and data analysis. In later chapters of this thesis (publication 6 and publication 7), GEPAT is used to analyze human microarray datasets and to integrate results from gene expression analysis with other datatypes. Gene expression and comparative genomic hybridization data from 71 Mantle Cell Lymphoma (MCL) patients was analyzed and allowed proposing a seven gene predictor which facilitates survival predictions for patients compared to existing predictors. In this study, it was shown that CGH data can be used for survival predictions. For the dataset of Diffuse Large B-cell lymphoma (DLBCL) patients, an improved survival predictor could be found based on the gene expression data. From the genes differentially expressed between long and short surviving MCL patients as well as for regulated genes of DLBCL patients, interaction networks could be set up. They point to differences in regulation for cell cycle and proliferation genes between patients with good and bad prognosis.