The generation of virtual patients plays a crucial role in quantitative systems pharmacology (QSP), particularly for simulating realistic patient populations in computational models. Allen et al. introduced a two-stage algorithm for generating virtual patients, where the first stage creates a “plausible population” with patient observables constrained within the clinical min-max range (1). The second stage refines this plausible population by sampling to match the clinical distribution. While this approach offers significant advancements, it faces limitations regarding the breadth of data it can support. The method was applied to cholesterol metabolism modeling using NHANES data, showcasing its utility.
Our goal is to build on this foundation by developing a platform that supports QSP modeling with flexible, scalable virtual patient workflows. This platform accommodates diverse likelihood functions, achieves substantial computational scalability, and provides a framework for continuous innovation in virtual patient generation and QSP applications.