Quantitative Systems Pharmacology (QSP) focuses on building models of pathophysiology and drug action. Due to the complexity of the relevant processes, much of the focus in the field is on constructing useful models through a combination of biophysics and data-driven techniques. Once one obtains such a model, however, the treatment design process remains non-trivial and is typically suboptimal, especially when combination therapies and patient-to-patient variation enter the analysis. In the traditional engineering disciplines, advances in control theory and reinforcement learning have allowed engineers to solve similar problems with formal safety guarantees. However, the complex and multi-scale nature of biological systems typically renders these algorithms computationally intractable. In this talk, I will discuss work we have done to bridge this gap, adapting these algorithms in the context of QSP to design optimally safe and effective drug regimens from pharmacokinetic/pharmacodynamic (PKPD) models.

The video of this presentation is available here.