Associate Director AstraZeneca Gaithersberg, Maryland, United States
Disclosure(s):
Khem Raj Ghusinga, PhD: No financial relationships to disclose
Quantitative Systems Pharmacology (QSP) models, frequently formulated as ordinary differential equations (ODEs), must account for biological variability arising from inter-individual differences. Standard approaches to incorporate this variability rely on sampling parameters and initial conditions from assumed distributions (e.g., lognormal) to propagate uncertainty through model simulations. However, achieving accurate representation of output distributions necessitates running a large number of simulations. Here, we report the results from a moment closure method for directly computing the time evolution of the statistical moments (i.e., mean, variance, etc.) of model states, offering an alternative that does not make assumptions about the underlying distributions and is potentially more efficient in terms of simulation time. We demonstrate that for linear systems, this technique provides exact analytical solutions for the moments. While nonlinear systems yield an infinite hierarchy of moment equations requiring approximation [1-2], employing well-established moment closure techniques provides results comparable to those from 10,000 sampled parameters of ODE output. Future work will compare the computational efficiency and accuracy of various moment closure methods against traditional sampling methods across a broader range of QSP models.
