Statistical issues in protein microarray analysis

Lianbo Yu, Purdue University

Abstract

Protein microarrays have emerged as a technology that has led to a new era of proteomics. Although protein microarray experiments allow the simultaneous study of hundreds or thousands of proteins in a high-throughput fashion, they also have a variety of experimental and statistical issues that need to be addressed. The purpose of this dissertation is to develop statistical methods to address the design and analysis issues, e.g., the selection of functionally consistent proteins and the identification of differentially expressed proteins. Since protein structures are usually not as stable as DNA molecules, the selection of functionally consistent proteins is an essential preliminary step in the fabrication of protein microarrays, Toward this end, a novel semi-nonparametric mixture model is proposed to classify functionally consistent proteins and functionally inconsistent proteins. The expression of functionally consistent proteins is then assessed under experimental conditions using microarray technology. Identifying differentially expressed proteins across different experimental conditions is the motivation for many protein microarray experiments. The many statistical approaches for identifying differentially expressed proteins range from parametric to nonparametric. Parametric methods often require strong distributional assumptions from the data, and while nonparametric methods are robust they may not be powerful. To balance the tradeoff between parametric and nonparametric approaches, a novel semi-nonparametric method for detecting differentially expressed proteins is proposed. Simulation results demonstrate the power of the semi-nonparametric approach as significantly higher than the nonparametric empirical Bayes approach while requiring minimal distributional assumptions.

Degree

Ph.D.

Advisors

Doerge, Purdue University.

Subject Area

Statistics|Biostatistics

COinS