Faster trial-like virtual populations from multiple datasets with Parallel Tempering
Renee Myers, QSP Programmer and Data Analyst, Rosa & Co. LLC
Renee Myers is a QSP programmer and data analyst for Rosa & Co LLC. She received her degree in Chemical Engineering from Oregon State University. Renee has been at Rosa for the past 5 years supporting modeling, simulation, and virtual population development efforts across multiple therapeutic areas. At Rosa, she has contributed extensively to novel methods for generation and analysis of virtual patients, several of which have been featured in past conferences, webinars, and recent publications.
Many quantitative systems pharmacology (QSP) programs stall when trial-like virtual populations that match multiple datasets are prohibitively time-consuming to generate. Teams are pushed toward smaller VPops, fewer scenarios, and uncomfortable compromises in trial design, dose selection, and combination strategy.
This webinar examines that constraint through a concrete oncology case study. A MAPK xenograft model under KRAS G12C and SHP2 inhibition was calibrated to several xenograft datasets and treatment arms in parallel and used to create a large number of virtual patients. In this setting, common VPop methods such as Simulated Annealing and Metropolis Hastings consumed substantial computation time and became harder to tune as the number of datasets increased. In contrast, Parallel Tempering generated roughly nine thousand plausible virtual patients across three treatment arms with an order of magnitude reduction in computation time while preserving goodness of fit.
Discussion will focus on what this unlocks for project teams in translational medicine, clinical pharmacology, and Pharmacometrics. Faster generation of diverse, trial-like VPops from multiple datasets supports more credible comparisons of Phase 2 designs, dose regimens, and combination strategies within existing timelines and compute budgets. The session compares Parallel Tempering to established methods of VPop generation and highlights practical patterns for using Parallel Tempering so variability, uncertainty, and multi-dataset evidence can be represented at the level decision makers expect.