" Design and Analysis of Large Scale Gene Expression Experiments and the Application to Angiogenesis and Blood Vessel Maturation"
-pdf

Abstract: High-throughput technologies such as DNA microarrays are changing the landscape of scientific research.  Instead of measuring RNA levels for one gene at a time, it is now possible to perform tens of thousands of simultaneous gene expression measurements.  Along with the obvious benefits afforded by this dramatic increase in scale, however, come numerous complications associated with data processing and analysis.  In this seminar I will present my doctoral research related to microarray experimental design and data analysis, agglomerative hierarchical clustering, and their application to evaluating differences in gene expression in an in-vivo model of angiogenesis and blood vessel network formation. 

To address many of the challenges associated with identifying differentially expressed genes in a microarray dataset, we have developed an experimental approach and supporting software that incorporate replication and a statistical linear model to account for known sources of variation.  This software, named CARMA (Computational Analysis of Replicate Measures for Arrays), performs an analysis of variance (ANOVA) on two-channel microarray datasets, in addition to all of the necessary data preprocessing steps including importing, transforming, and normalizing the raw data files. 

Once the differentially expressed genes have been identified, it is often desirable to group these genes based on their expression profile.  Numerous pattern recognition techniques, statistical approaches, and learning algorithms have been successfully applied to experimental microarray data, however evaluating the performance of each technique has proven difficult without knowledge of the true classifications within the data.  In order to evaluate the performance of hierarchical clustering algorithms, we developed software than generates simulated microarray datasets and implements 10 hierarchical clustering algorithms and 4 distance metrics.  Performance of each algorithm/distance metric combination was assessed based on their ability to recover the known clusters within the simulated datasets.

In an effort to improve our understanding of the cellular mechanisms regulating angiogenesis, blood vessel maturation, and vascular remodeling we utilized a mouse microvessel fragment model to study gene expression during the formation of a vascular network from small vessel fragments isolated from mouse periovarial and epididymal fat pads.  Over the course of 28 days, these small isolated fragments developed into a physiological microvascular network.  Analysis of gene expression at days 0, 3, 7, 14, 21, 28 revealed patterns of gene expression consistent with an initial angiogenesis phase followed by a maturation and network remodeling phase.