Code and data for: Composable probabilistic models can lower barriers to rigorous infectious disease modelling

Recent outbreaks of Ebola, COVID-19 and mpox, and routine surveillance of endemic pathogens such as influenza, have demonstrated the value of modelling for synthesising data to inform decision making. Effective models require integration of expert domain knowledge from multiple domains and outputs to be timely enough to inform policy yet current modelling approaches create barriers to meeting these goals. Methods used to synthesise available data broadly fall into approaches that chain separate models together, offering flexibility but losing information and potentially introducing bias, or rigorous joint models that are often monolithic and difficult to adapt. These barriers have prevented advances across multiple settings where models could have provided actionable insights. Composable models where components can be reused across different contexts and combined in various configurations whilst maintaining statistical rigour could address these limitations. We outline proposed requirements for a composable infectious disease modelling framework and present a proof of concept that addresses these requirements through composable epidemiological components built on Julia's type system and Turing.jl. We demonstrate a prototype R interface showing how such frameworks can bridge software ecosystems. Three case studies show how latent process components can be composed with epidemiological models to estimate time-varying reproduction numbers. The first replicates a COVID-19 analysis for South Korea using a renewal process. The second extends these components with reporting delays and day-of-week effects to replicate EpiNow2, a real-time nowcasting tool. The third replicates an ordinary differential equation model analysis of influenza outbreak data. We then discuss strengths, limitations, and alternative approaches. Our approach demonstrates promise for enabling interdisciplinary collaboration by lowering technical barriers for domain experts to contribute directly to model development. Future work is needed to solve remaining composability challenges, explore other options, expand the component library, and explore opportunities for large language model assisted model construction.

Keywords