Data-Driven Initialization of SParSE

Despite the ever-increasing affordability and availability of high performance computing platforms, computational analysis of stochastic biochemical systems remains an open problem. A recently developed event-based parameter estimation method, the stochastic parameter search for events (SParSE), is able to efficiently sample reaction rate parameter values that confer a userspecified target event with a given probability and error tolerance. Despite the substantial computational savings, the efficiency of SParSE can be further improved by intelligently generating new initial parameter sets based on previously computed trajectories. In this article, we propose a principled method which combines the efficiencies of SParSE with these geometric machine-learning methods to generate new initial parameters based on the previously collected data.