Nonparametric Hypothesis-Based Analysis of Microarrays for Comparative Phenotype Characterization

• Activity Number: 471
• Type: Contributed
• Date/Time: Thursday, August 7, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jeanne Kowalski*+ and Marianna L. Zahurak and Adam Mamelak and Daniel Sauder
• Companies: Johns Hopkins University and Johns Hopkins University and Johns Hopkins University and Johns Hopkins University
• Keywords: Genomics ; Microarrays ; Multivariate Analysis ; U-Statistics

Abstract

Microarray technology has made possible many research opportunities, such as the molecular characterization of cells. There remains, however, the formidable challenge of hypothesis testing for high-dimensional data from small samples. We propose a nonparametric, inference-based approach to comparing intensities from very few replicates on a few samples, within a high-dimensional setting. By introducing stochastic linear hypotheses for random vector pairs, we develop a series of related hypotheses in a high-dimensional space. One hypothesis examines whether gene intensities obtained from different sources may be summarized into a single-dimension statistic; another examines the data for differential expression and if rejected, separately examines under- from over-expression. All hypotheses are formulated and tested based on U-statistics and permutation methods. For describing candidate genes, we propose a bioinformatics approach that uses singular value decomposition and inner product concepts, in combination. The approach is applied to data from two-color arrays, where the problem of defining genes as potentially diagnostic for skin cancer is examined.