Generation of Virtual Populations for Quantitative Systems Pharmacology Through Advanced Sampling Methods

Area of Science:

• Quantitative Systems Pharmacology (QSP)
• Computational Biology
• Pharmacometrics

Background:

• Virtual population (Vpop) generation is crucial in QSP for simulating patient variability.
• High dimensionality and non-identifiability of QSP models pose significant challenges for Vpop generation.
• Existing methods like Metropolis-Hastings (MH) can struggle with complex parameter distributions.

Purpose of the Study:

• To evaluate the DREAM(ZS) algorithm for Vpop generation in QSP.
• To compare DREAM(ZS) performance against the traditional MH algorithm using a cholesterol metabolism model.
• To assess convergence, parametric diversity, and posterior coverage for complex biological models.

Main Methods:

• Utilized the DREAM(ZS) algorithm, a multi-chain adaptive Markov chain Monte Carlo (MCMC) method.
• Employed the Van De Pas model of cholesterol metabolism as a case study.
• Compared DREAM(ZS) with the single-chain MH algorithm, focusing on parameter space exploration and outcome correlations.

Main Results:

• DREAM(ZS) demonstrated superior parameter space exploration compared to MH.
• The algorithm effectively reduced boundary accumulation and restored parameter correlation structures.
• DREAM(ZS) utilizes an adaptive proposal mechanism and bias-corrected likelihood for improved sampling.

Conclusions:

• DREAM(ZS) offers a promising, user-friendly alternative for Vpop generation in QSP.
• The method enhances sampling efficiency in high-dimensional biological models.
• This contributes to improved in silico trial simulations and understanding inter-individual variability.