Your mission will be to develop and apply advanced likelihood-free statistical inference methods to improve the calibration of mechanistic epidemiological models. Indeed, mechanistic dynamic models are essential tools for anticipating and controlling the spread of infectious diseases at large scales. However, their performance critically depends on effective calibration using empirical data that are often complex, incomplete, and heterogeneous (epidemiological, serological, and genomic). There is currently a lack of robust methodologies to efficiently integrate such diverse data sources into complex epidemiological models. This project aims to develop likelihood-free statistical inference methods tailored to large-scale epidemiological models, using bovine viral diarrhoea virus (BVDV) as a case study. BVDV is an endemic disease characterized by silent spread through persistently infected animals, making it particularly suitable for methodological advances in data integration and transmission reconstruction.

You will be more specifically in charge of:

Development of inference methods: you will design and evaluate summary statistics for heterogeneous data integration, as well as implement and compare likelihood-free approaches (e.g., ABC-SMC, ABC-RF, Wasserstein-ABC, NPE) with respect to accuracy and computational efficiency.
Data requirements and surveillance design: you will identify the minimal quantity and quality of data required for reliable model calibration and assess how surveillance strategies (frequency, type, and coverage) impact inference performance.
Application to real-world data: you will develop a multi-strain mechanistic model for BVDV transmission in France, incorporating observation and sampling processes. You will also calibrate the model using the developed framework and reconstruct transmission patterns.