How to make more from exposure data? An integrated machine learning pipeline to predict pathogen exposure

Area of Science:

• Ecology
• Veterinary Science
• Computational Biology

Background:

• Predicting infectious disease dynamics in wildlife is crucial for management but challenging due to limited data and complex relationships.
• Existing methods often struggle with serological data from subsets of individuals and nonlinear ecological variables.

Purpose of the Study:

• To present a machine learning framework for constructing pathogen-risk models in wild animals.
• To automatically incorporate complex nonlinear relationships and handle missing data with minimal assumptions.
• To improve predictive performance and interpret disease risk insights using game theory.

Main Methods:

• Utilized statistical machine learning algorithms to build pathogen-risk models.
• Compared multiple machine learning algorithms in a unified environment to identify the best predictive model.
• Applied a game theory approach for enhanced interpretation of model results.
• Tested the framework on African lions infected with canine distemper virus (CDV) and feline parvovirus.

Main Results:

• The machine learning framework demonstrated superior predictive performance compared to traditional methods.
• Identified complex nonlinear patterns and interactions influencing pathogen exposure risk.
• Revealed specific risk factors, such as young age and low rainfall, for canine distemper virus exposure in lions.
• Provided new insights into disease risks in wild animal populations.

Conclusions:

• The developed framework offers a robust approach for predicting disease risk in wildlife populations.
• It effectively captures nonlinear patterns and complex variable interactions in ecological data.
• The methodology can be adapted for other ecological applications, including species distribution and diversity modeling.