|
Activity Number:
|
526
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #303994 |
|
Title:
|
Maximum Entropy Modeling for Gene Expression Microarray Data
|
|
Author(s):
|
Xue Lin*+ and Daniel Q. Naiman and Donald Geman
|
|
Companies:
|
FDA and Johns Hopkins University and Johns Hopkins University
|
|
Address:
|
1401 Rockville Pike, Center for Biologics Evaluation and Research, Rockville, MD, 20852,
|
|
Keywords:
|
Maximum entropy modeling ; gene expression microarray ; classification ; breast cancer
|
|
Abstract:
|
Maximum entropy modeling (MaxEnt) is a general approach in which the criterion for fitting a distribution to data is to maximize entropy subject to constraints on certain observed statistics. We apply this idea to modeling gene expression microarray data, using binary comparison statistics, and apply it to distinguishing between two cancer phenotypes. First, a small number k of gene pairs are selected based on maximizing a score which measures the extent of the reversal of expression across two phenotypes: the score is |Pr(A>B| class 1) - Pr (A>B|class 2)|, where A and B are the expression values of two genes. Then we build phenotype-specific models over permutations of size 2k and use the likelihood ratio test to predict the label of a new gene expression profile. We apply MaxEnt to two breast cancer data sets and demonstrate high accuracy in predicting BRCA1 mutations and ER status.
|
