Mixed-Effects Cox Models for Gene Set Analysis

• Activity Number: 673
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Terry Therneau*+ and Marianne Huebner
• Companies: Mayo Clinic and Mayo Clinic
• Keywords: survival analysis ; mixed effects ; gene sets

Abstract

Gene set analysis (GSA) methods take advantage of the fact that biological phenomena occur through interactions of multiple genes in functional relationships. Most GSA algorithms have been based on univariate tests statistics that are aggregated for a gene set score (bottom-up), or combining expression levels within gene sets into a single covariate (top-down). For time we propose to directly model the genes in a mixed effects Cox model. The gene set coefficients are treated as random effects with correlation coefficient r. As r -> 1, then the coefficients are forced to be identical (top-down model), ordinary shrinkage happens when r->0. In simulation results the approach is competitive with other methods, and has the advantage that covariates fit naturally into the framework.