Multi-scale Spatio-temporal Modeling, Learning, and Inference Methods for Complex Infectious Disease Systems

"Understanding epidemiological characteristics and spatio-temporal transmission patterns of infectious diseases plays essential roles in developing effective and efficient disease intervention strategies. However, the epidemic dynamics are extraordinarily complex, as they are influenced by various impact factors at the macroscopic level and the molecular evolution of pathogens at the microscopic level. Moreover, due to underreporting and misreporting of cases, the transmission dynamics of many contagious diseases are only partially observable, making the common practice of disease intervention inefficient. Therefore, there is a pressing need to model, learn, and infer the underlying disease transmission by making full use of observational data from a variety of sources and over a wide range of scales.

First, human mobility and their demographic structure play critical roles in formulating social contact patterns, which directly affect the disease transmission across various spatio-temporal scales. Therefore, how to model the epidemic dynamics from heterogeneous observational data at varying granularities is the first challenge that we aim to address. Second, in addition to the direct impact factors, there are also indirect factors such as climate change and socio-economic status, whose impact on the epidemic has yet to be explored. Recently, extensive machine learning methods have been developed to characterize the dependencies between factors and epidemic dynamics in a purely data-driven manner. However, how to integrate disease-specific knowledge into the learning process to enhance the model accuracy and interpretability is seldom investigated. Third, the evolutionary analysis of viral pathogens can provide additional information for epidemic inference from genetic data. Therefore, how to simultaneously infer both epidemiological and evolutionary characteristics of viral pathogens remains great challenges.

Accordingly, in this project, we aim to develop a unified computational approach, which is composed of three interrelated components: (i) A multi-granularity spatio-temporal transmission model based on human mobility and demographic structure; (ii) A hybrid data-driven epidemic-informed machine learning method that integrates various direct and indirect factors with disease-specific knowledge; (iii) A phylodynamic inference method that unifies molecular evolution and epidemic dynamics in structured populations. We will evaluate the developed approach by carrying out case studies on two typical infectious diseases, COVID-19 and AIVs, as globally encountered.

In summary, this project will offer novel computational solutions to the fundamental challenges of heterogeneous data modelling, machine learning, and statistical inference, and provide practical tools for AI and data science-enabled disease intervention and control."