Network based analysis of functional genomic data and variability of a human cell line

The measurement of gene expression levels using microarray technology is applied in many genomic and systems biological studies nowadays. In comparison to microarrays, the analysis of the metabolome and proteome is still limited regarding the number of measurable elements. Different studies indicate that the integration of a priori knowledge into the transcriptomic data analysis can improve the resulting interpretations of the cell metabolism regarding their reliability. This a priori information can be given, e. g., by the Gene Ontology terms or metabolic networks representing the cellular metabolism. A better understanding of the metabolism can also be obtained by an integrated analysis of the transcriptome, proteome and metabolome. In this work, two different tasks in the field of transcriptomics were addressed. First, a methodology was developed for the analysis of transcriptomic data using a metabolic network as framework for the analysis. The algorithm's performance was evaluated by its application to a simplified artificial network. The algorithm was also verified for its regularity. Furthermore, the robustness of the methodology was tested by varying a parameter of the underlying equations and the number of available transcripts in the analysis. The application of the developed method on yeast transcriptomic datasets from literature revealed amino acid pathways to be mainly affected during amino acid and nitrogen starvation conditions. Under the inuence of heat, the lipid metabolism was identified to play a major role in the adaption of the cells to the altered temperature. In the analysis of transcriptomic data of the yeast cell cycle, the application of the method predicted a correlation of the lipid metabolism with the G1 phase of the cell cycle. As the second task being addressed in this work the transcriptomic data of a biological replicate from a batch culture of the human cell line AGE1. HN were analyzed. The gene expression over time showed little similarity in the replicates regarding Pearson correlation coefficient and Spearman's rank correlation coefficient calculations, whereas the proteomic and metabolomic data showed significant similarities. The application of singular value decomposition on the transcriptomic data revealed similar patterns that accounted for > 60% of the information present in both datasets. Also the underlying gene-gene connection network showed a significant overlap. The overlapping network clusters mainly enriched into biological functions that are positively or negatively related to cell growth. Whereas the expression time courses of housekeeping genes showed no accordance in both datasets, these genes were directly connected to each other to 98% in each dataset. The decreasing degree of similarity from the metabolome to the transcriptome data as well as the preserved patterns and the gene-gene connection network were exemplified with a simplified controller scheme. The findings of this work indicate multiplicity of complex regulation on the level of gene expression that ensures similar resulting phenotypes in a repeated experiment while the transcript levels differ consistently. This integrated study of different omics data strongly suggests that individual gene expression time courses should be evaluated with great caution. Instead, gene expression time courses should be studied by comparing correlating genes, which is the gene-gene connection network, or by the analysis of global patterns within the datasets.