William Gillespie, Ph.D. presented at the Midwest Biopharmaceutical Statistics Workshop on May 15, 2018. See presentation here.
Stan (http://mc-stan.org/) is a widely used, open-source, probabilistic programming language
and Bayesian inference engine. Its primary inference engine is the NUTS algorithm, an HMC
sampler with adaptive adjustment of tuning parameters. NUTS is usually more efficient than the Metropolis-Hasting and Gibbs samplers, particularly for complex hierarchical models often used in pharmacometrics (PMX) applications.
Torsten (https://github.com/metrumresearchgroup/Torsten) is a library of Stan functions specific to PMX applications. Current Torsten functions provide convenient methods for specifying compartmental PK and PD models and schedules of events like dosing and observations. I will present examples of the use of Torsten/Stan for popPK and popPKPD applications. I will discuss the pros and cons of the use of Stan and Torsten relative to alternatives such as BUGS variants (WinBUGS, OpenBUGS, JAGS) and NONMEM. Stan and Torsten are the subjects of very active development programs, so I will close with a glimpse of things to come.