|
Activity Number:
|
382
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #307277 |
|
Title:
|
On Comparing the Clustering of Regression Models Method with K-Means Clustering
|
|
Author(s):
|
Li-Xuan Qin*+ and Steve Self
|
|
Companies:
|
Memorial Sloan-Kettering Cancer Center and University of Washington
|
|
Address:
|
307 E. 63rd Street, New York, NY, 10021,
|
|
Keywords:
|
clustering ; regression ; microarray ; gene clustering
|
|
Abstract:
|
Gene clustering is a common question addressed with microarray data. Previous methods, such as K-means clustering and the multivariate normal mixture model, base the clustering directly on the observed measurements. The clustering of regression models (CORM) method (Qin and Self 2006) bases the clustering of genes on their relationship to covariates and explicitly models different sources of variations. In this paper we discuss connections and differences between CORM and K-means clustering. We show that CORM tends to seek a partition of genes that has stable cluster centers across samples. Simulation results show that CORM outperforms K-means and an extended K-means when the assumed regression model is true and is robust to certain model misspecifications. We also use a microarray dataset to demonstrate a scenario where only CORM is appropriate.
|
