Developing better modelling inference tools to inform disease control for bovine Tuberculosis using epidemiological and pathogen genetic information.

Quantitative models are useful tools for projecting the outcome of disease control options, and therefore choosing between them. The epidemic of bovine Tuberculosis (TB) has generated a wealth of data which can be exploited to generate detailed predictive process models and evaluate their performance. Recently, the exploitation of pathogen sequence data has had a transformative impact on our understanding of epidemic diseases. In the context of mathematical modelling, the detailed representation of transmission pathways can greatly improve our ability to infer the values of model parameters that allows the models to recreate key characteristics of observed epidemics. Many of these methods have been developed for of rapidly evolving viruses with consistent evolutionary clocks, infecting a single host species. However, there remains a need to develop more general methods to infer transmission pathways in multi-host systems. A critical issue is that observations on all relevant host populations are often unbalanced, with data on one or more important hosts difficult to obtain. Recently we have used a simulation-based approach for considering the transmission of TB in Irish cattle and badgers, and identifies important epidemiological properties, despite the absence of any observations on the badger populations or infection in the badgers however these approaches need to be validated across different scenarios, and tested in scenarios where data across both host species are available. Further, while our approximate approach has demonstrated the ability to select between different badger contribution scenarios, the approach remains to be validated to make it useful across different scenarios. In parallel, we have also developed likelihood-based approaches for the simpler problem of FMD transmission in a single host system, as well as for the epidemiological analysis of an intensively studied badger epidemic.

In this project, we shall generate a suite of scenarios (endemic vs. epidemic, persistent in each population, only one population, or only in the two together) and different contact network relationships, to identify signals for transmission across the different scenarios, and propose new metrics for solving the underlying problems. We shall test these outcomes, we shall use extant datasets for M. bovis transmission with balanced cattle and badger information and very different transmission patterns. We shall consider two critical aspects of this process - first, by comparing the approximate and full likelihood methods we develop, we shall ask if the metrics in the approximate method are adequate for characterising the epidemic (sufficiently to the overall objective of modelling control) and second, if the model adequate for describing the processes relevant to choosing between disease control options. In the 1st part, we shall compare model outputs using the existing fitting approaches to the real data on disease outbreaks, and use this to develop recommendations of more relevant metrics (and using these in model fitting). In the 2nd, we shall propose up to three different model processes and structures based on epidemiological insight (e.g. the potential role of supershedders, or variation in the ability of the standard test to detect infected cattle), use these to generate synthetic datasets which will be fitted to the baseline model using the different metrics proposed in part one, and then demonstrate the relative ability of the model fitted to these different metrics to fit the synthetic data and predict to outcome of control.

Therefore we shall both developing methods to consider in detail generalisable multi-host phylodynamic models, & address key issues for the management of an important disease problem, thereby facilitating more tailored approaches to control of bTB and other multi-host diseases.

We shall develop new methods, approaches & guidance to inform endemic and epidemic disease scenarios in multi-host systems via integrated analysis of disease dynamics and pathogen sequence data. We shall test and showcase these tools by application to state of the art phylogenetic data from the cattle-badger Tuberculosis (TB) system in GB. There is an explosion in the availability of pathogen sequence data from disease outbreaks. However, standard methods for longitudinal data analysis that account for disease dynamics do not exploit pathogen sequence data, & standard approaches for pathogen sequence data analysis only abstractly account for the dynamics of disease. Recent work including by the project partners has demonstrated the potential of such integrated analysis in the context of a rapidly evolving pathogen in a single host species (e.g. foot and mouth disease virus in cattle). We shall create the next generation of such spatial phylogenetic epidemic tools to enable for the first time their routine use in the analysis of fast or slowly evolving pathogens in multi-host systems. To achieve this we shall:

1. Create dynamic spatial phylodynamic models for multi-host pathogen systems focussing on TB in badgers and cattle.
2. Develop and assess, under model misspecification, novel Bayesian inference frameworks based on: recently developed importance distributions combined with particle filters; explicit likelihood-based data-augmentation Markov chain Monte Carlo and model assessment tools; and Approximate Bayesian Computation summary statistics.
3. Identify typical signatures for a range of data availabilities and scenarios, from emerging epidemic outbreaks to long term endemic disease, and characterise their potential to inform disease control.
4. Apply the developed methods to data on the badger-cattle-TB system to analyse both core endemic areas and those at the leading edge of epidemic spread, and thus provide insight into this important disease problem