Multiscale Infectious Disease Dynamics Inference (MIDDI)

Standard methods for inferring infectious disease transmission typically rely on a single scale of data-such as population-level case counts or binary individual-level test results-and assume simple infection dynamics within hosts [Hay 2021]. However, such methods are often inadequate for answering important research and policy questions that require a more detailed mechanistic understanding of disease transmission, including: the influence of pathogen load or antibody level on an individual's infectiousness or susceptibility; the role of different infection sources in sustaining transmission; the extent to which transmission is driven by superspreading events or highly infectious individuals; how individual-level immune responses shape population-level transmission dynamics. Answering these questions requires integrating diverse and granular data types-such as longitudinal molecular or serological measurements, genomic data, and spatial contact information-into mechanistic individual-level transmission models. This, in turn, demands more complex inference methods capable of handling high-dimensional missing data, latent processes (such as unobserved infections and infection times), and strong dependencies between parameters and data. Data-augmentation MCMC [Jewell 2009a] is one such method, but existing implementations are typically designed for specific diseases, model structures and data types [Jewell 2009b, Spencer 2015, Lau 2015, Salje 2018, Chapman 2020]. As a result, there is a lack of general-purpose inference frameworks that can be applied flexibly to multiscale datasets integrating within- and between-host dynamics. This limits the accuracy and depth to which we can understand and predict the spread of different diseases. The overall aim of this project is to fill this gap by developing a Bayesian framework for performing inference on multiscale (within- and between-host) infectious disease data that can be applied to a range of diseases. This will be achieved by: Developing a modelling and inference framework for inferring individual-level transmission that incorporates pathogen genomic data and multiple forms of within-host test data; Extending this framework to more complex within-host dynamics (e.g. changing pathogen load or antibody levels) and coupling within- and between-host infection dynamics; Developing new inference tools for the multiscale infection dynamics framework to accelerate inference; Applying the framework to multiple diseases to provide deeper insight into their transmission and control. We will fit multiscale models of infectious disease transmission to data on visceral leishmaniasis, AMR E. coli and K. pneumoniae infection, malaria, and COVID-19. This will allow us to simultaneously reconstruct transmission trees, infer unobserved infections, and estimate individual-level time-varying infectiousness-tasks that are beyond the reach of conventional inference approaches. In turn, this will enable us to make more accurate and detailed predictions of disease spread and to inform improvements in control interventions, potentially leading to better health for large numbers of individuals.