Using Machine Learning Surrogate Modeling for Faster QSP VP-Cohort Generation
Christina Friedrich, PhD; Jérémy Huard
Chief Engineer, PhysioPD, Rosa & Co, LLC; Senior Application Engineer, MathWorks
Wednesday February 16, 2022, 12:00 to 1:00 pm EST
Register for free at https://www.rosaandco.com/webinars
Abstract:
Virtual patients (VPs) are widely used within QSP modeling to explore the
impact of variability and uncertainty on clinical response. In one method of
generating VPs, parameters are sampled from a distribution, protocols are
simulated, and the possible VP is either accepted or rejected based on
constraints on model output behavior, such as achieving reasonable responses to
clinical protocols. The approach works but can be inefficient, i.e., the vast
majority of model runs typically do not result in valid VPs.
Machine learning (ML) surrogate models offer an opportunity to greatly improve
the efficiency of VP creation. Surrogate models are trained using the full QSP
model to discriminate between parameter combinations that result in feasible
VPs vs. those that do not. Once the surrogate models are developed, parameter
combinations can be pre-screened rapidly, and the overwhelming majority of
pre-vetted combinations result in valid VPs when tested in the original QSP
model.
In this webinar, Rosa and MathWorks will present this novel workflow and give a
case study example using a psoriasis disease QSP model from the Rosa
PhysioPD(tm) practice and the MATLAB® Regression Learner app to select and
optimize the surrogate models. The VPs generated by the surrogate modeling
approach are statistically similar to VPs generated using only the original QSP
model. We conclude with comparisons of the relative efficiency of the methods,
and ideas for expansion of the use of this and other ML methods in QSP modeling.