| The analysis and interpretation of the large data sets produced by high-throughput biological experiments are among the most important and challenging problems in science today. The goal of this thesis is to demonstrate the utility of a phenomenological approach to the study of statistics in general, with a particular focus on understanding the results of such experiments.;The Renormalization Group (RG) has been a popular tool in physics for the construction of phenomenological theories. RG may be interpreted as a search for stability, and it is in this form that we make use of it here. We show that an RG stability argument (against addition of new data) may be used to 'derive' standard statistical quantifiers such as the mean and variance. The utility of this principle is also demonstrated in the context of more general quantifiers. In particular, we show how it may be used to guide choices in the development of a novel dimensional reduction scheme.;We have proposed a method, called the ICS Survey, that uses these ideas in the realm of multiple experiments. The ICS survey is a data driven method that exploits the differences between experiments by using them to perturb the system and identify stable parts. By doing so, it successfully identifies the dominant processes (in those experiments) and the genes involved in them, thereby solving some of the more vexing problems faced by exploratory dimensional reduction methods. It is also one of the few methods attempting to make use of the information contained in inter-experiment variability.;We have also discussed various methodological issues faced in the analysis of high-throughput experiments. In particular, novel methods for noise removal and for visualization are presented. |
