|
Activity Number:
|
81
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 12, 2002 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section*
|
| Abstract - #301387 |
|
Title:
|
Expression-level-Dependent Standardization of Gene Expression Differences
|
|
Author(s):
|
Alexandre Bureau*+ and Josee Dupuis
|
|
Affiliation(s):
|
Genome Therapeutics Corporation and Genome Therapeutics Corporation
|
|
Address:
|
100 Beaver Street, Waltham, Massachusetts, 02453,
|
|
Keywords:
|
variance function ; nonparametric regression ; microarrays ; comparative experiment ; heteroscedasticity
|
|
Abstract:
|
Microarray experiments to compare gene expression levels between different conditions often include too few replicates to reliably estimate gene-specific null distributions of summary statistics. As a result, many methods proposed to analyze such experiments rely on the assumption that summary statistics are identically distributed for all genes. The observation that the variability in expression measurements strongly depends on the level of expression contradicts that assumption. When at least two replicates are available, we propose to standardize measures of expression differences by an estimate of standard deviation parameterized as a function of the expression level. We investigated parametric and nonparametric functions. We found that local polynomial regression of squared residuals provides a better fit than parametric variance functions, as evidenced by a greater homogeneity of the distribution of the residuals after standardization. The methods are applied to both oligonucleotide and spotted cDNA microarray datasets, and we will discuss how the method adapts to various experimental designs.
|
