Fall Seminar Series - November 10, 2005
University of Minnesota
School of Statistics
College of Liberal Arts

Signal Quality Measurements for cDNA Microarray Data

Tracy L. Bergemann
Division of Biostatistics
University of Minnesota

Thursday, November 10, 2005
3:30 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:00 PM, 300 Ford Hall

Abstract

Concerns about the reliability of transcription rates estimates derived from microarray data inspires ongoing research into measurement error in these experiments. Error happens at both the technical level within the lab and the experimental level. In this paper, we are concerned with measuring spot-specific error.

This paper outlines two different approaches to quantify the reliability of spot-specific intensity estimates. In both cases, the spatial correlation between pixels and its impact on spot quality is accounted for. The first method is a straightforward parametric estimate of within-spot variance that assumes a Gaussian distribution and accounts for spatial correlation via an overdispersion factor. The second method employs a non-parametric quality estimate referred to throughout as the mean square prediction error (MSPE). The MSPE first smooths a pixel region and then measures the difference between actual pixel values and the smoother.

Both methods herein are compared for real and simulated data to assess numerical characteristics and the ability to describe poor spot quality. Further, we examine the use of spot quality estimates to increase efficiency in downstream analysis.