Does the Why Lead to the How to Build Virtual Populations?

(*Jointly authored by Richard Allen & Ted Rieger) The virtual population concept in QSP models arises from a necessity to make predictions with an assessment of both variability and uncertainty in response. The power of virtual populations is to predict the response of a population including dynamics that might not have been foreseen based on calibration data. Therefore, robust predictions with QSP models requires careful construction of virtual populations. This perspective directly informs the methods which are most suitable for the generation of virtual patients and populations, and how novel approaches (for example, incorporating AI/ML into the workflow) can accelerate their development. In this talk we will review the philosophy and necessity of building virtual populations for QSP models, discuss which approaches best align with these, and provide concrete examples of the predictive power of robust virtual populations.