Resolving outbreak dynamics using approximate Bayesian computation for stochastic birth–death models

Abstract

Earlier research has suggested that approximate Bayesian computation (ABC) makes it possible to fit simulator-based intractable birth-death models to investigate communicable disease outbreak dynamics with accuracy comparable to that of exact Bayesian methods. However, recent findings have indicated that key parameters, such as the reproductive number , may remain poorly identifiable with these models. Here we show R that this identifiability issue can be resolved by taking into account disease-specific characteristics of the transmission process in closer detail. Using tuberculosis (TB) in the San Francisco Bay area as a case study, we consider a model that generates genotype data from a mixture of three stochastic processes, each with its own distinct dynamics and clear epidemiological interpretation.

ABC of the future

“…Simulator‐based inference is well suited to infectious disease epidemiology. For example, it has been used to resolve the outbreak dynamics of stochastic birth‐death‐mutation models (Lintusaari et al , 2019), and to infer the transmission dynamics of the Ebola haemorrhagic fever outbreak in 1995 in the Democratic Republic of Congo (McKinley et al , 2009) and the COVID‐19 pandemic (Chinazzi et al , 2020). In this case study, we demonstrate the application of simulator models to gain insight into the epidemic caused by the emergence of the Ebola virus in West Africa in 2014 as reported by the WHO Ebola Response Team (Team WER) (WHO Ebola Response Team, 2014) to infer the basic reproduction number , that is, the mean value of secondary infections caused by an infectee when no control measures are in place.…”

“…For example, Bayesian optimisation for likelihood‐free inference (BOLFI), has been shown in several benchmark examples to speed up ABC inference by 3–4 orders of magnitude (Gutmann & Corander, 2016), and multiple successful applications of it beyond typical benchmarks used in the statistical literature have emerged. These include applications in very diverse research fields, such as inverse reinforcement learning for cognitive user interface models (Kangasrääsiö et al , 2017), brain task interleaving modeling (Gebhardt et al , 2020) and more general computational models of cognition (Kangasrääsiö et al , 2019), perturbation modeling and selection in bacterial populations (Corander et al , 2017), direct dark matter detection (Simola et al , 2019), pathogen outbreak modeling (Lintusaari et al , 2019), sound source localisation (Forbes et al , 2021), passenger flow estimation in airports (Ebert et al , 2021), and the modeling of the genetic components that control the transmissibility of pathogens (Shen et al , 2019). To inspire further methodological development, software engineering and dissemination of ABC and other LFI methods, we present here an array of real applications and discuss both the benefits and challenges that lie ahead for this exciting subfield of statistics.…”