A Bayesian Nonparametric Approach to PK/PGx Studies

• Activity Number: 52
• Type: Invited
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Abel Rodriguez*+
• Companies: University of California at Santa Cruz
• Keywords: Pharmacokinetic ; Pharmacogenetic ; Gaussian process ; Dirichlet process

Abstract

Pharmacokinetic/Pharmacogenetic (PK/PGx) studies explore the time course of drug concentrations resulting from a particular dosing regimen, and how genetic variation affects such outcomes. Statistical models for PK/PGx studies typically use nonlinear regression models centered around the solution of a system of ordinary differential equations (ODEs) to describe the blood-concentration of the drug for each subject in the sample. Subject-specific parameters are assumed for the ODEs, and information is shared across patients by modeling these parameters using generalized linear mixed models. In practice, this approach is simple but unsatisfactory; for example, variability across subjects is often not well explained by simply changing the parameters of the ODE. This talk explores the use nonparametric regression models based on Gaussian processes to capture functional variability across subjects. The models are centered around the solution of traditional PK models based on ODEs, but allow greater flexibility in modeling genetic and subject-specific variability. In order to determine what genetic markers influence the functional response, we also introduce a new hierarchical prior