----------------------------------------------------------------------- BIOINFORMATICS COLLOQUIUM School of Computational Sciences George Mason University ----------------------------------------------------------------------- Two New Tools for Analyzing and Understanding Gene Expression Data Karen Schlauch GMU Tuesday, April 6, 2004 4:30 pm Verizon Auditorium, Prince William Campus The analysis of microarray data is a significant challenge for any researcher both in terms of finding appropriate ways to reduce the data set to components that carry useful information and for finding patterns within the data that convey biologically important interpretations. When an experiment is performed across a set of time states, an additional level of complexity is imposed as patterns of expression variation for one gene as well as expression patterns across genes must be considered. Thus the parallel assay of many thousands of data points, not all of which are independent, across a set of temporal states provides an interesting platform for applying statistical analyses and the testing the construction of models. Two techniques developed to analyze gene expression across temporal states will be presented. The first is a novel graph-theoretic approach for constructing putative functional network models that suggest hypotheses about functions of unknown genes. An innovative distance metric provides a measure of similarity between any pair of genes in a more biologically grounded manner than distance metrics used in common hierarchical clustering methods. Using these similarity relations, a bi-directional graph is generated. Gene "clusters" can be detected within the structure of the graphUs connected components. This technique has been applied to several experiments of promising results. The second is an approach to building a model of gene interactions over a set of temporal states. Boolean networks are often used as a first approximation to model complex systems. Using the gene expression data derived from a set of temporal experimental states, a Boolean network representing gene interactions is constructed. This model, however, is far from unique. In order to reduce the space of relevant Boolean representations, we employ mathematical and statistical techniques to find a best model of the selected gene set. ---------------------------------------------------------------------- Refreshments are served at 4:00 pm. Find the schedule and directions at http://www.binf.gmu.edu/colloq.html