Generation of Virtual Populations for Quantitative Systems Pharmacology Through Advanced Sampling Methods
by Miriam Schirru, Tristan Brier, Maxime Petit, Didier Zugaj, Pierre-Olivier Tremblay, and Fahima Nekka
Read the paper
Quantitative systems pharmacology (QSP) models are designed to capture biological complexity and variability, and their utility increasingly relies on the generation of virtual populations (Vpop). Generating high-quality Vpop remains challenging due to nonlinear, high-dimensional, and often non-identifiable nature of QSP models. We adapted the multi-chain DREAM(ZS) MCMC algorithm to QSP and compared it against Metropolis-Hastings using the van de Pas cholesterol model. DREAM(ZS) achieved broader parameter space coverage, restored correlation structures, and generated more diverse Vpop without sacrificing model fit. It offers a flexible framework for QSP, not only for steady-state but also for future applications involving dynamic outputs.
Overview of the DREAM(ZS) algorithm applied to quantitative systems pharmacology modeling, showing its workflow, comparison with Metropolis-Hastings, and improvements in virtual population generation.
