|
Activity Number:
|
384
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #300394 |
|
Title:
|
Cluster Analysis for Continuous Data with an Excess of Zeros
|
|
Author(s):
|
Kimberly Siegmund*+
|
|
Companies:
|
University of Southern California
|
|
Address:
|
1540 Alcazar Street, CHP-220, Los Angeles, CA, 90089,
|
|
Keywords:
|
class discovery ; cluster analysis ; mixture models ; gene expression
|
|
Abstract:
|
Abnormal DNA methylation is typical in cancer. Since DNA methylation profiles vary across tumor types and subtypes, it is believed that clustering DNA methylation profiles may uncover novel disease subgroups. DNA methylation data obtained using the MethyLight technology is quantitative with an excess of zeros. For a region of DNA, MethyLight measures the occurrence of fully methylated alleles. For many samples it finds none while for others it finds variable levels. We introduce a Bernoulli-lognormal mixture model for clustering continuous data with an excess of zeros and compare it to standard model-based clustering methods for discrete data and for continuous data. In a simulation study we find the Bernoulli-lognormal mixture model has the lowest misclassification error rate compared to competing approaches. We illustrate the methods using DNA methylation profiles from a study of lung cancer cell lines. The Bernoulli-lognormal mixture model has the lowest cross-validation error for distinguishing lung cancer subtype (non-small cell vs. small cell) and allocates samples to classes with the lowest uncertainty.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2004 program |
