Sensitivity Analysis in the Face of Rare Events

Molecular motors and other complex nonequilibrium systems are controlled by large sets of design parameters, and optimizing those parameters requires computing sensitivities -- derivatives of dynamical observables with respect to the parameters. When the system's dynamics involves rare events, both the observable and its sensitivity are difficult to estimate from direct simulation. We present a practical computational pipeline that addresses both challenges by combining importance sampling with a Markov state model (MSM). The MSM separately captures the slow, rare-event dynamics and the fast, local dynamics, and the chain rule connects those two pieces to yield an efficient sensitivity estimator. An iterative reweighting procedure based on the RiteWeight algorithm substantially reduces approximation errors from the MSM coarse-graining. We validate the approach on diffusion in the Müller-Brown potential, where the sensitivity of a transition rate to landscape parameters can be computed exactly. We then use sensitivies to optimize the directional bias of a particle-based model of a catalysis-driven molecular motor.