Supervised learning and prediction of spatial epidemics.

Area of Science:

• Epidemiology
• Computational Biology
• Plant Pathology

Background:

• Parameter estimation for infectious disease models is computationally demanding.
• Supervised learning offers an alternative to traditional model fitting methods like Bayesian Markov chain Monte Carlo.
• Previous work successfully applied classification to predict disease model characteristics from epidemic curves.

Purpose of the Study:

• To extend supervised learning inference to inherently spatial infectious disease models.
• To compare the efficacy of global versus spatially stratified epidemic curves for classifier training.
• To analyze a real-world plant disease dataset using the developed methods.

Main Methods:

• Simulating epidemics from various spatial infectious disease models.
• Building classifiers using epidemic curve data (global and spatially stratified).
• Applying the classification approach to a tomato spotted wilt virus dataset.

Main Results:

• Spatially stratified curves provide more accurate inference for spatial models compared to global curves.
• The classification approach successfully identified underlying spatial disease dynamics.
• The method was validated on simulated data and applied to a real-world epidemic.

Conclusions:

• Supervised learning, particularly with spatially stratified data, is a viable and efficient method for inferring parameters in spatial infectious disease models.
• This approach offers a computationally efficient alternative to traditional parameter estimation techniques.
• The findings have implications for understanding and managing spatially explicit disease outbreaks.