|
Activity Number:
|
102
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #310017 |
|
Title:
|
Conjugate Hierarchical Modeling of the Error Variance in Tests for Differential Gene Expression
|
|
Author(s):
|
Jason Osborne*+
|
|
Companies:
|
North Carolina State University
|
|
Address:
|
2501 Founders drive, Raleigh, NC, 27695-8203,
|
|
Keywords:
|
microarray ; shrinkage estimation ; optimal discovery procedure ; regularization
|
|
Abstract:
|
A conjugate mixing distribution for the error variance is proposed to model inhomogeneity of gene variances in microarray experiments. The likelihood obtained by integrating over this conjugate is maximized to estimate mixing parameters. The predictive distribution of the error sums of squares from linear gene models is more flexible than the usual chi-square distribution and can improve fit. The estimated conditional mean of the variance shrinks the gene-specific variances towards an estimate that pools over all genes, with weights on gene-specific error mean squares proportional to the number of replicate arrays, thereby sensibly borrowing strength across genes. Simulations are used to investigate improved sensitivity of the resulting regularized F-tests as well as performance when the shrinkage technique is used for estimation of Storey's optimal discovery procedure.
|
