|
Activity Number:
|
211
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #304044 |
|
Title:
|
Weighted Kernel Fisher Discriminant Analysis for Integrating Genomic and Clinical Data with Application to Cancer Prediction
|
|
Author(s):
|
Jemila S. Hamid*+ and Celia M.T. Greenwood and Joseph Beyene
|
|
Companies:
|
Hospital for Sick Children and University of Toronto and University of Toronto
|
|
Address:
|
555 University Ave, Toronto, ON, M5G 1x8, Canada
|
|
Keywords:
|
data integration ; Fisher discriminant analysis ; kernels ; weighted kernel discriminant analysis
|
|
Abstract:
|
We propose a method for integrating heterogeneous data sets. Each data set is represented as a kernel matrix. We perform kernel Fisher discriminant analysis (KFDA) using each of the data sets and define a weight based on classification error. The kernel matrices are combined in a weighted fashion where the weight for each data set represents its relative importance. KFDA on the combined kernel is then performed to classify individuals into subclasses. The method can be applied to combine two or more heterogeneous data sets. We illustrate our approach by integrating gene expression and clinical data sets with the aim of improving breast cancer prediction. For our illustrative data, the method resulted in a weight of 0.4859 and 0.5141 for the clinical and gene expression data, respectively, indicating that both data sets are equally important. wKFDA is being performed on the combined data.
|
