Stochastic Mixed Effects Models with Particle Markov Chain Monte Carlo in Pharmacokinetics/Pharmacodynamics

Abstract:

Pharmacokinetic/pharmacodynamic (PK/PD) models heavily rely on non-linear ordinary differential equations (ODEs), which are most often encompassed in non-linear mixed effects population models to account for inter-individual variations. Since the mid-2000s PK/PD models have been taken to the next level with the replacement of ODEs by stochastic differential equations (SDEs), to consider uncertainty in the biological dynamic system itself – hence allowing intraindividual variation. Parameter calibration in this new, more elaborate class of models calls for new inferential tools. Recent pioneering attempts include frequentist Maximum Likelihood via Laplace approximations and Bayesian posterior sampling via Gibbs sampler. However, they both demand strong practical restrictions on the class of SDE considered. In this account of

our early-stage work, after an introduction to harmacokinetic/pharmacodynamic applications, we present here how the recent Particle Markov Chain Monte Carlo (PMCMC) algorithms allow to extend Bayesian inference to the broadest class of models, conveniently weakening those tractability restrictions. This is a joint work with with Arnaud Doucet (UBC) and Gareth Peters (UNSW).

About the speaker: Dr Julien Cornebise is currently a Postdoctoral Fellow in the Laboratory for Computational Intelligence (LCI) at the Department of Computer Science of the University of British Columbia (Vancouver, Canada), under the supervision of Pr. Arnaud Doucet. He was previously a Postdoctoral Fellow at the Statistical and Applied Mathematical Sciences Institute (SAMSI) (North Carolina, USA). His main research interests are Sequential Monte Carlo (SMC) methods, Adaptive Monte-Carlo methods and Stochastic optimization.