2020/01/05 by Thomas Catanach, Catanach, Thomas A., Huy D. Vo +3 · 1 citation
Computer Science · Decision Sciences · Engineering · #62C10 #62F15 #65C05 #65C40 #92C42 #Bayesian Modeling and Causal Inference #Computation (stat.CO) #FOS: Computer and information sciences #Fault Detection and Control Systems #G.3 #J.3 #Probabilistic and Robust Engineering Design
paper · pdf · doi:10.48550/arxiv.2001.01373
openalex publication_date 2020/01/05 · openalex created_date 2022/07/22 · openalex updated_date 2026/07/28
Stochastic reaction network models are often used to explain and predict the\ndynamics of gene regulation in single cells. These models usually involve\nseveral parameters, such as the kinetic rates of chemical reactions, that are\nnot directly measurable and must be inferred from experimental data. Bayesian\ninference provides a rigorous probabilistic framework for identifying these\nparameters by finding a posterior parameter distribution that captures their\nuncertainty. Traditional computational methods for solving inference problems\nsuch as Markov Chain Monte Carlo methods based on classical Metropolis-Hastings\nalgorithm involve numerous serial evaluations of the likelihood function, which\nin turn requires expensive forward solutions of the chemical master equation\n(CME). We propose an alternative approach based on a multifidelity extension of\nthe Sequential Tempered Markov Chain Monte Carlo (ST-MCMC) sampler. This\nalgorithm is built upon Sequential Monte Carlo and solves the Bayesian\ninference problem by decomposing it into a sequence of efficiently solved\nsubproblems that gradually increase model fidelity and the influence of the\nobserved data. We reformulate the finite state projection (FSP) algorithm, a\nwell-known method for solving the CME, to produce a hierarchy of surrogate\nmaster equations to be used in this multifidelity scheme. To determine the\nappropriate fidelity, we introduce a novel information-theoretic criteria that\nseeks to extract the most information about the ultimate Bayesian posterior\nfrom each model in the hierarchy without inducing significant bias. This novel\nsampling scheme is tested with high performance computing resources using\nbiologically relevant problems.\n