Temporal Activation Profiles of Gene Sets for the Analysis of Gene Expression Time Series

Abstract

This thesis focuses on the analysis of high dimensional quantitative gene expression data generated by high throughput time series experiments. The statistical analysis is challenged by the typically large number of genes compared to the number of observations (small number of replicates at a small number of time points). The key strategy facing these problems is the analysis on the level of a priory defined gene sets. The statistical power is increased and the interpretation of the results is much easier due to the biological knowledge, which originates from the gene set definition.

Five algorithms associating each considered gene set with a temporal activation profile are presented against the background of typically used strategies from the literature to analyze gene expression time series. Two extensive simulation studies are conducted to compare the methods and to evaluate the proposed smoothing techniques.

The estimation of temporal activation profiles for gene sets from the Gene Ontology, KEGG, Biocarta, Reactome and Biocyc definitions is applied on four mouse time series experiments. The resulting profiles consist of one symbol per time point (+ for enrichment with up regulated genes, - for down regulated genes and o for no enrichment). The comparison with the original contributions in the literature reveals both a high conformity with the previously registered findings and a large proportion of new insights. Hence, the proposed algorithms turn out to be a helpful tool for an exploratory analysis of gene sets on gene expression time series.