|
Activity Number:
|
487
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 7, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #302098 |
|
Title:
|
Disease Prognosis by Multiple-Gene Classifier with Pair Information
|
|
Author(s):
|
Shuyan (Sabrina) Wan*+ and Xiang Yu and Peggy Wong
|
|
Companies:
|
Merck Research Laboratories and Merck Reserach Laboratories and Merck Research Laboratories
|
|
Address:
|
RY34-A316, Rahway, NJ, 07065,
|
|
Keywords:
|
Random Forest ; Boosting ; disease prognosis ; paired data ; biomarker
|
|
Abstract:
|
Molecular profiling techniques for disease prognosis have been actively applied in biomedical research recently. When clinical information is available, it is attractive for medical researchers to match patients from two disease (or treatment) groups into pairs to increase the power to detect differentially expressed genes. For such design, it is necessary to take into account of the correlation within pairs and between genes to build a multiple-gene classifier. Thus we proposed ensemble-learning methods with modification such as Random Forest (RF) and Boosting methods to build the classifier. Simulation studies showed that both the RF and Boosting methods were able to build a composite biomarker and predict the incoming patient's group with certain accuracy. The RF method was also applied to a blinded dataset in cardiovascular disease and the results were promising.
|
