Speaker: Andrew Golightly (Durham University)

Title: Accelerating Bayesian inference for stochastic epidemic models using incidence data

Abstract: This work considers the case of performing Bayesian inference for stochastic epidemic compartment models, using incomplete time course data consisting of incidence counts that are either the number of new infections or removals in time intervals of fixed length. The most natural Markov jump process representation of the model is eschewed for reasons of computational efficiency, and replaced by a stochastic differential equation representation. This is further approximated to give a tractable Gaussian process, that is, the linear noise approximation (LNA). Unless the observation model linking the LNA to data is both linear and Gaussian, the observed data likelihood remains intractable. Unlike previous approaches that use the LNA in this setting, two approaches for marginalising over the latent process are considered: a correlated pseudo-marginal method and analytic marginalisation via a Gaussian approximation of the noise model. These approaches are compared using synthetic data with the best performing method applied to real data consisting of removal incidence of Oak Processionary moth nests in Richmond Park, London.  

2nd Seminar: 10.30-11am UK time

Speaker: Henrik Häggström (Chalmers University)

Title: Simulation-based inference for stochastic nonlinear mixed-effects models with applications in systems biology

Abstract: We propose a novel methodology for Bayesian inference in hierarchical mixed-effects models. By building on our work [1], we construct a simulation-based inference (SBI) framework that is highly scalable, where amortized approximations to the likelihood and the parameters posterior are first obtained, and these are rapidly refined for each individual dataset, to ultimately approximate the parameters posterior across many individuals. Unlike the current state-of-art SBI methods, which use neural networks, our approximations are expressed via Gaussian mixture models, leading to easily trainable, parsimonious yet expressive surrogate models of both the likelihood function and the posterior distribution. The methodology is exemplified via stochastic differential equation mixed-effects models to describe translation kinetics after mRNA transfection, however the methodology is general and can accommodate other types of stochastic and deterministic models. We compare our approximate inference with exact pseudomarginal inference and show that our methodology is fast and competitive.