Prediction via Sparse Kernel PCA

• Activity Number: 215
• Type: Invited
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Tianxi Cai*+
• Companies: Harvard School of Public Health
• Keywords: kernel machine ; risk prediction ; principal component analysis ; shrinkage methods

Abstract

Accurate risk assessment can have a great impact in public health. The standard approach to constructing risk prediction rules is to assume a linear effect and fit models such as the GLM. However, when markers relate to the phenotype simultaneously via a complex structure, prediction rules based on such linearity assumptions may not be effective. To overcome such difficulties, we employ a kernel machine regression framework and estimate prediction rules that incorporate potential non-linear and interactive effects of a set of markers, such as genes within a pathway. To achieve an optimal trade-off between the complexity of the model and the variability in the estimation, we consider sparse kernel principal component regression that leads to a reduced effective degree of freedom. Simulation studies suggest that the proposed procedures work well in finite sample.