Developing and Testing Methods for Microarray Data Analysis Using an Artificial Life Framework

Microarray technology has resulted in large sets of gene expression data. Using these data to derive knowledge about the underlying mechanisms that control gene expression dynamics has become an important challenge. Adequate models of the fundamental principles of gene regulation, such as Artificial Life models of regulatory networks, are pivotal for progress in this area. In this contribution, we present a framework for simulating microarray gene expression experiments. Within this framework, artificial regulatory networks with a simple regulon structure are generated. Simulated expression profiles are obtained from these networks under a series of different environmental conditions. The expression profiles show a complex diversity. Consequently, success in using hierarchical clustering to detect groups of genes which form a regulon proves to depend strongly on the method which is used to quantify similarity between expression profiles. When measurements are noisy, even clusters of identically regulated genes are surprisingly difficult to detect. Finally, we suggest cluster support, a method based on overlaying multiple clustering trees, to find out which clusters in a tree are biologically significant.