Bridging the Gap Between Mechanistic and Phenomenological Models of Cell Survival via the Manifold Boundary Approximation Method (MBAM)

We apply the Manifold Boundary Approximation Method (MBAM) to systematically compress the parameter space of Mechanistic DNA Repair and Survival (MEDRAS), a complex mechanistic radiobiological model for cell survival in the presence of radiation. We first apply MBAM to a simplified analytic version of MEDRAS, where the cell survival probability can be determined in closed-form. This explicit representation renders the Fisher Information Matrix (FIM) calculable through automatic differentiation techniques. The complete set of fifteen parameters in the analytic MEDRAS model is observed to display a sloppy hierarchy of parameter sensitivity, such that essentially all characteristics of the dose-response are encapsulated in a reduced three-parameter model, which has a natural interpretation in terms of the phenomenological target theory of radiobiology, but with parameters that can now be directly related to more fundamental mechanistic parameters of DNA damage and misrepair. We conclude by discussing efforts to extend our approach to more general stochastic MEDRAS simulations, where the cell survival probability does not usually have a closed-form representation.