|
Activity Number:
|
382
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #307075 |
|
Title:
|
A Divisive Method via Multivariate Hypothesis Testing for Clustering Gene Expression Patterns
|
|
Author(s):
|
Haiyan Wang*+
|
|
Companies:
|
Kansas State University
|
|
Address:
|
108 E. Dickens Hall, Manhattan, KS, 66506,
|
|
Keywords:
|
longitudinal data ; high dimensional data ; clustering ; hypothesis testing ; microarray data ; time course data
|
|
Abstract:
|
Linear method to cluster genes based on expression patterns often need data preprocessing to achieve reasonable results though data preprocessing could actually corrupt the original data and introduce spurious temporal behavior (Taguchi & Oono 2005). Here we present a divisive method to cluster genes according to expression patterns. The measure of dissimilarity between groups is given by the p value from a nonparametric multivariate hypothesis testing for no group differences all at time points. The test statistic uses overall ranks of expressions so that the analysis is invariant to monotone transformations of data. Asymptotic distribution of the test statistic is obtained and used to obtain the p values. Simulation shows that the procedure can extract significant patterns with low error without any supervision or preprocessing even in case of small number of observations per cycle.
|
