Year: 2019
Type: conference-paper
Abstract: Stochastic models of biochemical reaction networks are often more realistic descriptions of cellular processes over deterministic counterparts when small populations of certain chemical species are considered. Realistic applications of these models are computationally intensive since analytical solutions to the chemical master equation are rarely available. Multilevel Monte Carlo (MLMC) methods have been demonstrated to be very effective for the acceleration of expectation calculations with respect to the forwards problem. The statistical inference of reaction rate parameters, however, is an even more computationally intensive task that often relies upon likelihood-free methods, such as approximate Bayesian computation (ABC). We investigate a modified ABC approach that is based on and application of MLMC applied directly to rejection sampling; Our method constructs an approximation of the posterior distribution function through a telescoping summation of biased estimators using a decreasing sequence of acceptance thresholds. We briefly review the concept of MLMC for the forwards problem, discuss challenges specific to the inference problem, then demonstrate the effectiveness of our MLMC method for ABC using several stochastic models of biochemical reaction networks.
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2980902095