Development of Virtual Population for a Quantitative Systems Pharmacology model
Kapil Gadkar (1), Saroja Ramanujan (1)
(1) Development Sciences, Genentech
Objectives: Quantitative systems pharmacology (QSP) is an emerging discipline that focuses on analysis of the dynamic interactions between drug(s) and biological systems to understand the behavior of the system as a whole, as opposed to isolated behavior of individual components (van der Graaf & Benson 2011). Given the integrative nature of the approach, the corresponding mathematical and computational models frequently incorporate a wide range of data from in-vitro, preclinical, and clinical studies. This work describes a 4-stage approach for virtual population development suited for QSP models.
Methods: The first step involves identification and exploration of model parameters to generate numerous potential virtual patients; this exploration is based typically on in-vitro and preclinical data. In the second stage, simulations of the potential virtual patients are compared to clinical data of interest to select “relevant” virtual subjects. In the third stage, the selected virtual patients are validated against clinical data not used in the selection stage. A fourth model refinement stage is required for the scenario in which the validation criteria are not met in stage 3. Refinement could involve modifications of the model structure to include additional details of the biological process or could be limited to additional parameter exploration and virtual patient generation. The virtual population developed through this process represents both potential variability in the underlying biology and clinical variability observed in the “real” population. Each patient in the virtual population is assigned a prevalence weight [1], such that the weighted virtual population reproduces key statistical properties of clinical population data. These properties could include means and distributions of clinical measurements and readouts at baseline and/or in response to therapies, as well as correlations between multiple measurements. Thus, the prevalence weight of each individual virtual patient represents its contribution to statistical properties of the virtual population.
Results: The virtual population development approach is described for a QSP model of LDL cholesterol lowering by anti-PCK9.
Conclusions: The above systematic methodology of data integration and variability exploration into a virtual population is well suited for in-silico research in QSP models.
References
[1] Schmidt, B. J., F. P. Casey, et al. (2013). "Alternate virtual populations elucidate the type I interferon signature predictive of the response to rituximab in rheumatoid arthritis." BMC Bioinformatics 14: 221