Abstract:

This dissertation presents computational relevantly exhaustive demonstration (CRED) standards as an approach to the combination of differing standards. Standards and computational tools can play an important role in scientific research, and specifically for Affymetrix DNA microarray experiments, by performing computational evidence generation that can identify and reduce bias introduced by data analysis methods.

Using computational techniques, we demonstrate that the results from microarray experiments can vary greatly depending on the specific quantitation and normalization standards used, despite the fact that these methods are intended to accomplish the same task. Further, we show that microarray researchers, likely unaware of these unpredictable instances of bias, routinely do not correct for these sources of variability, leading to a lack of robustness in research findings. Additionally, microarray research publications often do not provide sufficient information to replicate the experimental findings.

To remedy this situation we present BAGeL, a new CRED standard for obtaining differentially expressed gene lists from microarray data by aggregating output from multiple traditional standards. We demonstrate that this approach highlights disagreement among different standards for detecting differential gene expression, and provides much more selective gene lists than these standards.

We also introduce LAffy, an automated parallel-processing Affymetrix DNA microarray data analysis package that provides standardized data analysis tools for researchers and facilitates the development of CRED standards. LAffy's automation eliminates the need for researchers, who are generally not trained in computer science, to spend large amounts of time identifying, locating, learning, installing, and executing many data analysis tools. In addition to increasing the accuracy of research results, LAffy increases efficiency and lowers the cost of conducting microarray experiments. LAffy also includes an implementation of our new BAGeL standard.

Further, we introduce MIAME CRED, a new CRED standard extension of the MIAME standard for microarray experiment information sharing and data analysis that can substantially increase the quality and transparency of microarray research findings.

Through developing these new computational techniques, we highlight the benefits that can be achieved through the coordinated use of CRED standards.