Integrative Data Mining and Meta Analysis of Disease-Specific Large-Scale Genomic,Transcriptomic and Proteomic Data

Abstract

During the past decades, large-scale microarray technologies have been applied to the field of genomics, transcriptomics and proteomics. DNA microarrays and mass spectrometry have been used as tools for identifying changes in gene- and protein expression and genomic alterations that can be linked to various stages of tumor development. Although these technologies have generated a deluge of data, bioinformatic algorithms still need to be improved to advance the understanding of many biological fundamental questions. In particular, most bioinformatic strategies are optimized for one of these technologies and only allow for an one dimensional view on the biological question. Within this thesis a bioinformatic tool was developed that combines the multidimensional information that can be obtained when analysing genomic, transcriptomic and proteomic data in an integrative manner. Neuroblastoma is a malignant pediatric tumor of the nervous system. The tumor is characterized by aberration patterns that correlate with patient outcome. aCGH (array comparative genomic hybridization) and DNA-microrarray gene expression analysis were choosen as appropriate methods to analyse the impact of DNA copy number variations on gene expression in 81 neuroblastoma samples. Within this thesis a novel bioinformatic strategy was used which identifies chromosomal aberrations that influence the expression of genes located at the same (cis-effects) and also at different (trans-effects) chromosomal positions in neuroblastoma. Sample specific cis-effects were identified for the paired data by a probe-matching procedure, gene expression discretization and a correlation score in combination with one-dimensional hierarchical clustering. The graphical representation revealed that tumors with an amplification of the oncogene MYCN had a gain of chromosome 17 whereas genes in cis-position were downregulated. Simultaneously, a loss of chromosome 1 and a downregulation of the corresponding genes hint towards a crossrelationship between chromosome 17 and 1. A Bayesian network (BN) as representation of joint probability distributions was adopted to detect neuroblastoma specific cis- and trans-effects. The strength of association between aCGH and gene expression data was represented by markov blankets, which where build up by mutual information. This gave rise to a graphical network that linked DNA copy number changes with genes and also gene-gene interactions. This method found chromosomal aberrations on 11q and 17q to have a major impact on neuroblastoma. A prominent trans-effect was identified by a gain of 17q.23.2 and an upregulation of CPT1B which is located at 22.q13.33. Further, to identify the effects of gene expression changes on the protein expression the bioinformatic tool was expanded to enable an integration of mass spectrometry and DNA-microrarray data of a set of 53 patients after lung transplantation. The tool was applied for early diagnosis of the Bronchiolitis Obliterans Syndrome (BOS) which occurs often in the second year after lung transplantation and leads to a repulsion of the lung transplant. Gene expression profiles were translated into virtual spectra and linked to their potential mass spectrometry peak. The correlation score between the virtual and real spectra did not exhibit significant patterns in relation to BOS. However, the metaanalysis approach resulted in 15 genes that could not be found in the seperate analysis of the two data types such as INSL4, CCL26 and FXYD3. These genes constitute potential biomarkers for the detection of BOS