| In the past, an experiment involving a pair wise comparison normally involved one or a few dependant variables. Now, 1000s of dependent variables can be measured simultaneously in a single experiment, be it detecting genes via a microarray experiment, sequencing genomes, or detecting microbial species based on DNA fragments using molecular techniques. How we analyze such large collections of data will be a major scientific focus over the next decade. Statistical methods that were once acceptable for comparing a few conditions are being revised to handle 1000.s of experiments. Molecular biology techniques that explored 1 gene or species have evolved and are now capable of generating complex datasets requiring new strategies and ways of thinking in order to discover biologically meaningful results. The central theme of this dissertation is to develop strategies that deal with a number of issues that are present in these large scale datasets. In chapter 1, I describe a microarray analytical method that can be applied to low replicate experiments. In chapter's 2-4, the focus is how to best analyze data from ARISA (a PCR based molecular method for rapidly generating a finger print of microbial diversity). Chapter 2 focuses on qualifying ARISA data so that data will best represent its biological source, prior to further analysis. Chapter 3 focuses on how to best compare ARISA profiles to one another. Chapter 4 focuses on developing a software tool that implements the data processing and clustering strategies from chapter's 2 and 3. The findings described herein provide the scientific community with improved analytical strategies in both the microarray and ARISA research areas. |
