No One-Size-Fits-All: A Multi-Model Evaluation of Optimization-Based and Surrogate-Assisted Virtual Population Generation in QSP

Motivation & Background

• Virtual populations (VPops) represent uncertainty and inter-
  individual phenotypic variability in QSP models, supporting clinical
  decision-making [1]
• Widely used approaches following Allen et al. [2] generate large
  plausible populations (PPops) and select VPops to match target
  clinical distributions
• Such methods are typically evaluated on single models,
  assuming transferability
• As QSP model complexity increases, PPop generation becomes a
  major computational bottleneck